人気: データ更新日:2026年11月19日

カテゴリ

インターネット1,416
コミュニケーション328
音楽・動画938
画像・デザイン563
オフィス・仕事効率化1,179
開発ツール3,284
AIツール608
セキュリティ・プライバシー408
システムツール3,187
ゲーム543
教育・科学242
ネットワーク・IT管理293

Anchorsの旧バージョン・全バージョン一覧

Spec-first continuity framework for AI-assisted development

Anchorsのバージョンを、2026年8月28日から最新の0.2.3まで227件掲載しています。いずれも提供元の公式インストーラーにリンクしています。

2026: 227 227 26
バージョン日付サイズファイル
0.2.3最新版 2026年11月19日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.2.3
- A product's changelog lies between its own tags, per branch. With a versioning: block, anchors changelog reads a product's releases between its own tags, ordered by its scheme, each headed by the version its tag records (1.10.0, not app-v1.10.0); another product's tags do not split them. A branch rule takes release: true (its changelog lists the releases) and changelog: true (where --write writes); a branch that does not release lists the candidate, what came after the latest release, headed by the version its strategy computes. A product whose rules name no such branch releases, and writes, on every branch. --product and --branch pick them. (CLGCM, VRSNN)
- A product declares its changelog outputs. versioning.<product>.changelog.outputs lists files or directories, each with its path, mode, template and format; --write writes every one of them. A product with no outputs keeps the changelog: block. (VRSNN, CLGCM)
- A stage that releases writes the tag and the GitHub Release. A stage with release: <product> writes, once it passes and before its record, the product's tag on its sha with the version the sha carries, then the tag's Release with the changelog since the release before. A tag already on the sha is a rerun; on another sha, the same version was released twice, and the stage is recorded as failed. The workflow grants contents: write, and pipeline-wired diverges when it does not. (RLSTA, PCSPA, PWSPP)
- New gate changelog-valid. Declared with scope: project, it confronts every output, whoever wrote it: markers that read as versions, newest first, none twice, the unreleased block on top, a block for every tag and a tag for every block, and the block of a release unchanged against HEAD unless the commit message waives the gate. The output's format: adds the heading with its version and its tag's date, the per_version file names, required sections, no empty section, max_chars, and required: changed while the sources hold a version not released yet. (CHVLA)
With versioning: declared, anchors changelog now heads each release by its version rather than by its tag, and its markers name the version: a changelog file written before by tag names has its blocks written again. To use the new keys, install this binary first: an older one refuses them.
Changelog
- 10be940 docs(release): v0.2.3 notes
- bd6d2a5 feat(changelog): a product's changelog lies between its own tags, per branch
- ddded73 feat(gate): changelog-valid confronts what a product's changelog outputs hold
- 73ed014 feat(pipeline): a stage that releases writes its product's tag and Release
0.2.2 2026年11月19日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.2.2
- A stage asks of the version its sha carries. A new requirement in a stage's requires, version: { product, fresh, listed, same_as_passed, expected }, is answered at the sha, never in the working tree. fresh: the version is above every release, the product's tags wherever they are reachable and what the product's new released: lists; a version still in review is not released, so resubmitting it passes. listed: { where: {...} }: a file gives the version an entry with those fields, by default the file and map released: declares. same_as_passed: <stage>: the version is the one the commit that passed that stage carried, so production releases what passed review. expected: the version is the one the branch's strategy computes; manual asks it be fresh, and none refuses. (VRQRA, PCSPP)
- A stage asks that the fixes came back. contains: <branch> fetches that branch from origin and holds when its head is an ancestor of the sha; when it is not, it names how many commits are missing and the first five. (VRQRA)
- A branch may bound its versions. A rule of versioning.<product>.branches takes range:, an explicit range (1.0.x) or from_base, the release the branch was cut from at its last part. Every version: requirement and anchors version bump refuse a version outside it, and anchors version shows it. (VRSNN, VRSNI)
- released: lists releases the tags do not record. A file, the map whose keys are versions, and the fields an entry must hold, for the releases made before the project wrote tags. (VRSNN)
- anchors branch check confronts the pull request's head. Besides the flow, it answers the version: and contains: requirements of every stage of the base, on the head commit (--head-sha, or the event's pull_request.head.sha), so a refusal shows on the pull request and not later in the stage. same_as_passed stays with the stage. (BCSBR)
The tags must be in the CI's checkout (fetch-depth: 0) for fresh and range: to see them. To use the new keys, install this binary first: an older one refuses them.
Changelog
- ec6b05b docs(release): v0.2.2 notes
- 78d1ffe feat(pipeline): a stage asks of the version its sha carries, and of the fixes it contains
0.2.1 2026年11月18日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.2.1
- A product declares how its version is written, ordered and moved. The new versioning: block maps each product to a scheme, the files that hold its version, its tag and the strategy each branch decides the next version by. The schemes are semver, calver with a declared format, number, and pattern, a regular expression whose groups are the version's numbers. The strategies are manual, increment of a chosen part, semantic (from the commits since the latest release) and none. A branch overrides the product's default only where it differs, and the first pattern it matches wins. A Version: footer on a commit forces the next version. Nothing assumes a language, a package manager or a store: the files, the paths, the tag and the rules are the project's. A project without the block sees no change. (VRSNE, VRSNN)
- anchors version shows where a product stands. It prints the version each source holds, the latest release (the highest tag of the product reachable from HEAD, by the scheme's order), the strategy the branch resolves to and the next version with its reason. --json prints the same with the commits since the release classified, for a project's own script. anchors version bump writes the strategy's version, a raised part or a given version into every source, changing only the version's bytes, and refuses a version that does not move forward. (VRSNI)
- New gate version-forward. Declared with scope: project, it confronts the version the sources hold with the one at HEAD: kept or raised passes; lower, the same rank spelled differently, unreadable, or sources that disagree fail. (VRFRA)
To use versioning: in a project, install this binary first: an older one refuses the unknown key.
Changelog
- 4c2afa8 docs(release): v0.2.1 notes
- 7d3d4d2 feat(gate): version-forward confronts a product's version with the one at HEAD
- ee268cf feat(version): products declare how their version is written, ordered and moved
0.2.0 2026年11月18日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.2.0
- A new start for the repository. The history of this repository was restarted from a single commit. The command line behaves as it did in v0.1.422: nothing a project declares in anchors.yaml changes, and the format of the map stays 8.
- Every release before v0.2.0 is retracted. go.mod retracts [v0.0.0, v0.1.422], and those releases are no longer published here. A project whose min_version names one of them, or whose pipeline downloads one by its tag, moves to min_version: 0.2.0.
- The design notes left the repository. The DESIGN-*.md and FINDINGS-*.md files were working notes, not part of the product, and the references to them were removed from the code, the specs and the site.
Changelog
- 3385705 chore(release): v0.2.0 starts the repository over
0.1.422 2026年11月16日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.422
- check --changed no longer climbs past an import of types alone. In MIF, a hotfix that changed one sentence of a legal text module reached 195 nodes. Validation climbed from the module to the screen and then to the navigator. From there it went through every screen that imports the navigator only for its param types, and up to each of their specs. That blocked rule-implemented on 63 specs and evidence-fresh on 18 flows the change could not touch. A parent tied only by an import the project marks with no side effect on the tests is now still confronted, but the climb stops there, as anchors impact has done since v0.1.412. On the same tree, the scope went from 1127 to 146 nodes. (IMANM-B10)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 364935d docs(release): v0.1.422 notes
- 049fc22 fix(check): the scope stops climbing at a parent that imports only types
0.1.421 2026年11月16日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.421
A demand's plan can now be changed, both automatically and by hand.
- A plan moved by the commit takes its demand along. In MIF, a plan moved with git mv from plans/review/ once it was approved left its demand naming a file that no longer existed, and no command could change it. Now, when the commit moves the plan of a demand, the pre-commit hook points the demand at the new path in anchors.progress.md and says so. It stages the progress file when that file has no other unstaged changes; otherwise it asks you to stage it. A failure warns and does not block. (MPSYN-B11)
- anchors progress replan <demand> <plan>. It points a demand at another plan file: a moved one, or a new one. The plan must be an existing file. The demand's title, stages and tasks stay. (LDCML-B11, DMLDD-B16)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 314402e feat(progress): a moved plan takes its demand along, and progress replan points it by hand
0.1.420 2026年11月16日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.420
- A merge GitHub does not associate with its commit is still found. The branch flow and the pipeline's flow: requirement look up the pull request a guarded branch's commit was merged from by asking commits/{sha}/pulls. That association is an index GitHub fills apart from the merge, and it can lag or never come. In MIF, a pull request merged into production while GitHub was returning 502s answered merged: true with its merge commit, but the association stayed [] for more than fifteen minutes: pipeline can-run production-release refused the commit as "the merge of no pull request", and the release stopped at the guard. When no associated pull request was merged as the commit, the recently closed pull requests are now read for the one whose merge commit it is. Reported by MIF. (BCSBR-B04)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 4c338b8 fix(branch): a merge GitHub does not associate with its commit is found among closed PRs
0.1.419 2026年11月16日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.419
- env-declared can ask a unit's spec for what its dependencies read: follow_dependencies and provided. Where each deployed unit gets only the variables declared for it, such as a function per handler, a variable that a shared helper reads is part of that unit's deploy contract. Two new options on the gate entry control this. (ENVDC-B08)
- follow_dependencies: [<layer>…] names the layers of code where the gate also follows depends-on, transitively and past no import of types. In those layers, a variable the unit's dependencies read is asked of the unit's spec, as if the unit's own code read it.
- provided: [<glob>…] names the variables the platform gives every unit (AWS_*, NODE_ENV), which no spec is asked to declare.
- name: env-declared
check: env-declared
follow_dependencies: [handler, admin-handler]
provided: ["AWS_*", "NODE_ENV", "SENTRY_*", "APP_ENV"]
Off by default: without follow_dependencies, nothing changes. The options are scoped because of measurements in MIF. Switched on for every layer, 421 specs were charged, the app's build variables and the runtime's own among them. Scoped to the handlers, with the runtime's and the shared variables provided, 107 specs were charged, mostly with table names their helpers read.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 8cb9626 feat(gate): env-declared follows the dependencies of the layers the project names
0.1.418 2026年11月16日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.418
- The files after the flags join --changed. In anchors check --changed a b c, the shell hands b and c to the command as arguments. Since v0.1.409 check and verify refused them. Now they join the list, as a comma-separated list would, so --changed $(cat list) works. (CGPCH-B101)
- An argument that names no file of the project is still refused, before anything runs, and the error names it. A path read relative to the root or to where the command runs counts as a file of the project. Nothing given is dropped without a word. (CGPCH-E09, VPFVR-E06)
- In anchors test, an argument is a layer when one is declared by that name, and a file of the project otherwise. A file joins --changed. An argument that is neither is refused, and the error says what an argument may be. The hint added in v0.1.407 is gone, since the file is now simply taken. (STPRS-B18)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 1a578d3 feat(check): the files after the flags join --changed, in check, verify and test
0.1.417 2026年11月16日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.417
- A comment-only edit keeps the coverage fresh. A proof of execution already survived an edit that only touched comments. The line coverage did not: an inserted comment line moves the numbers of the lines below, and the map carried line-level signals only when every line stayed in place. In MIF's mutation campaign, each // Stryker disable marker asked for the workspace's whole suite to run again with coverage.
The map now records which lines of a file are code (kept_lines). When only comments changed, the coverage measured before is moved line by line to the new numbers and stays fresh: the lines each suite instrumented and covered, and the branches no suite took. The i-th line of code before is the i-th after. (EDSTD-B26, RPSCR-B63)
Nothing is moved when:
- a comment that steers the coverage tool changed (istanbul ignore, c8 ignore, pragma: no cover…);
- the map does not yet know the old layout, which happens until a file is built once by this version.
The mutation score is never moved: a comment may be a directive of the mutation tool, and the score is measured again.
- New key coverage_directives:. It lists the patterns of those comments. When unset, the common tools' directives apply: istanbul, c8, v8, coverage.py, gcov/lcov and tarpaulin. When set, the list replaces them. (CNFGO-B81)
The first anchors map build after the upgrade records the layout of every code file.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 8ce573f feat(mapx): a comment-only edit moves the coverage to the new line numbers
0.1.416 2026年11月16日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.416
- The commit hook no longer rewrites the tree's map with the commit's revisions. When the tree is ahead of the index, the hook stages the commit's own map, built from the index, and writes the map on disk back as the tree's. But it took what the commit's map needs from that same disk map, and carrying evidence moves the source's nodes to the commit's revisions. A file edited in the tree and not staged therefore went back to disk with the index's revision and the tree's line revision. Its coverage, measured on the tree, then read stale, and ingesting the lcov again did not clear it. In MIF this happened to 64 files whose only edits were // Stryker disable comments. The hook now reads the disk map twice, once for the merge and once to write back, so the map on disk keeps the tree's revisions and measurements. Only the dates the hook wrote move in both. Reported by MIF. (MPSYN-B10)
A map already affected is repaired by anchors map build, which takes each node's revisions from its file again, followed by ingesting the coverage once more.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 176cd10 fix(hooks): the hook no longer rewrites the tree's map with the commit's revisions
0.1.415 2026年11月15日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.415
- mutation-score leaves out the survivors the tool marks static. Stryker marks a mutant "static": true when it only runs while the module loads, such as a top-level const initializer. With coverageAnalysis: perTest the module is already imported when the mutant is switched on, so the mutation never takes effect and it comes out "Survived" with the suite green. ignoreStatic drops only the static mutants that no test covers. In MIF, 3 of a handler's 9 survivors were static, and applying one by hand broke its test at once.
- A static survivor is now counted apart: neither a survivor nor a mutant that ran, so it is out of the numerator and the denominator. It is kept on the map (mutants_static).
- The gate names the count in its detail, whatever the verdict. A file whose only survivors are static does not apply, and says so.
- A static mutant that was killed is still killed.
Reported by MIF. (MTINM-B11, SGINA-B35, INCHN-B44)
The score changes at the next mutation ingestion of each file.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- e9808e3 fix(mutation): a survivor the tool marks static is out of the score, counted apart
0.1.414 2026年11月15日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.414
- New verdict level: awaiting a decision (⏸). It is a choice registered on purpose that a person decides and no agent can clear. Before, these were counted as divergences (⚠) and sat in the pile of what to fix: 12 of some 440 in MIF. Now they have their own level:
- a ⏸ column in the table of gates and a line in the legend;
- a list after the findings, with what each one does, the gate, the target and who decides;
- a line under the verdict counting them.
They are not counted as divergences. (GTENG-B38, CGPCH-B100)
- open-questions-resolved: an identified open decision is now ⏸ and names who decides, read from the third cell of its row in the Open Decisions table. It still becomes a decision issue. An item with no code and a spec with no section stay ⚠, because they have a mechanical way out. (OPQSP-B09)
- doctrine-realized: a doctrine whose unrealized rules are all @TBD is now ⏸. It is declared, so it informs and never bars. (DCTRN-B12)
- severity: { awaiting: block | inform | ignore }. The new level takes a state like the others, per gate or as the project default. When unset, it blocks on a blocking gate and informs on an informative one. (CNFGO-B80)
Behaviour change: on a gate declared blocking: true, an open decision now bars the commit. Before, the default severity of a divergence (inform) won over the gate's own mark, and an open decision did not bar. The default gate the framework generates is not blocking and is unaffected. To keep open decisions listed without barring:
- name: open-questions-resolved
blocking: true
severity:
awaiting: inform
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 0214a1e feat(gate): awaiting verdict (⏸) — a choice registered on purpose that a person decides
0.1.413 2026年11月15日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.413
- layer-boundary matches the code, not its comments. A forbidden pattern is now matched against the file with the comments of its language blanked, keeping lines and offsets. In MIF, a JSDoc that named a color (#FFF) and a remark that cited an issue (#712) failed the "use a token" rule for hex colors. A rule whose target is a comment, such as one that forbids @ts-ignore or eslint-disable, declares in_comments: true and is matched against the whole content, as before. A language with no comment syntax declared is read whole. Reported by MIF. (LYBNL-B12)
- doctrine-not-duplicated measures a unit too small to weigh its own words. The similarity weights (IDF) need at least four rules. When a spec and the doctrine it realizes have fewer between them, the rules of the project's other specs and doctrines now join the corpus of the weights. The pair compared is still the spec's rule against the doctrine rule it realizes. In MIF, a spec of one rule realizing a doctrine of two stayed pending, and the only way out was to invent rules. Reported by MIF. (DCTRN-B34)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 427088f fix(gate): layer-boundary skips comments; doctrine-not-duplicated measures small units
0.1.412 2026年11月15日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.412
- anchors impact lists who depends on the files. Before, the "Propagates to" list followed every edge outward. A depends-on edge runs from the importer to the imported file, so from a screen the list reached what the screen uses. Through the navigator the screen imports, it reached every screen of the app: in MIF, 833 items "to redo" for one screen whose evidence closure holds five tests. The command now always goes to the side that depends, whatever the edge type:
- whoever imports the file or declares @dep of it;
- whoever reads what it writes, captures it or realizes it;
- for a spec, the code, feature and tests it rules.
- Levels. Without a flag, only the direct dependents show. --propagate indirect climbs one more level, and --propagate cascade climbs to the end of the chain. Each item past the first level shows the path that ties it to the file.
- Large branches. An item that ten or more nodes depend on says how many it pulls, so you can judge whether that branch belongs. --stop-at <glob> lists such an item and climbs no further from it.
- Types-only imports. A tie the project marks with no side effect on the tests, such as an import of types, is listed and marked, but the walk does not continue past it. Its tests are not added.
- Tests to run again. The list holds the tests reached and those whose evidence closure holds a reached file, the same closure evidence-fresh judges. They are grouped by the layer of the suites that run them.
- Other output. "Validates against" still lists the spec, feature and guides that govern the file. --format files (with --layer) prints the tests for a runner, and --format json prints everything. Several files can be asked about at once. (IMANM-B06–B09, MPCTI-B07–B09, MPCTI-E03)
- New command: anchors dependencies <file>.... It walks the same ties the other way: everything the files depend on. That means what they import or declare @dep of, the writers of what they read, what they capture or realize, and the specs, features and guides that rule them. It takes the same --propagate, --stop-at and --format flags. (DPNDS-B01–B02)
check --changed keeps the slice it had. Its impact radius is unchanged.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- fa023bd feat(impact): impact lists who depends on the files, by level; new dependencies command
0.1.411 2026年11月15日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.411
Two defects in anchors affected-tests, reported by MIF. Before, it returned nothing for a screen whose evidence closure holds five tests.
- A named file with no change since the last commit is read whole. The command compares each file with its committed version, so it can name the tests of the rules and regions that changed. Asked about a file right after the commit that changed it, it compared the file with itself, found nothing, and printed an empty list with no warning. Such a file now reaches every test whose evidence closure holds it. (AFTSF-B06)
- Test files whose titles cannot be read are still listed. The files format listed only files with a test title the dialect could read. Flows of another tool, such as a Maestro YAML that captures a screen, were dropped. The files format now lists them with the other tests. The title formats (lines, regex, json) name them separately on the error stream. (AFTSF-B07)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- ce1bbbe fix(affected-tests): an unchanged named file is read whole, and untitled test files are listed
0.1.410 2026年11月14日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.410
- anchors docs build compiles a page whose @anchors header sits above its marker. The compiler only writes a page that opens with its anchors:generated marker, so it never overwrites a page written by hand. The format 7 migration wrote the project's @anchors header at the top of every compiled page that carried a code. The marker was then no longer the first line, and the page read as handwritten. In MIF, 25 pages under docs/ were reported as "NOT generated — they exist in docs/ without the marker" and were not updated after 2026-10-04. The marker now counts when it follows an @anchors header (an HTML comment, or # lines in a YAML). The page is judged, compiled, pruned and staged by what follows the header, and the header is kept as it is. A page with a header and no marker is still treated as handwritten. Reported by MIF. (DTCDC-B25)
Run anchors docs build once to bring such pages up to date.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 2502887 fix(docs): a compiled page whose @anchors header sits above the marker is compiled again
0.1.409 2026年11月13日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.409
- check and verify refuse positional arguments. In anchors check --changed a b c, the flag takes a, and the shell passes b and c as positional arguments. The command accepted them and dropped them, then judged a alone and answered that every gate passed. In MIF, 1 of 1728 files was judged, and the check read green. Both commands now fail before judging anything. The error names the arguments and the two ways to list files: one per flag (--changed a --changed b), or a comma-separated list (--changed a,b,c). Reported by MIF. (CGPCH-E09, VPFVR-E06)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- b897fb4 fix(check): check and verify refuse positional arguments instead of dropping them
0.1.408 2026年11月13日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.408
- A spec whose codes are all proven again at its current revision reads fresh. A spec's proof is only as fresh as its oldest contributing suite. Proofs ingested by hand, under reports no suite writes again (out/fx.xml, t22.xml…), therefore kept a spec stale forever. In MIF, 15 specs read "stale test signal (the spec changed since the ingestion — reingest)" after the unit suite had proven every one of their codes at the current revision, and no run could clear them. An old suite now holds the spec back only for codes it alone proves. A code that no suite has proven at the current revision still keeps the spec stale. Reported by MIF. (SGINA-B34)
The fix applies to a spec's freshness when a suite that proves it is next ingested. A spec stuck this way clears with one run of its tests, for example anchors test --changed <spec>.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 06c6e1a fix(ingest): a spec whose codes are all proven again at its current rev reads fresh
0.1.407 2026年11月13日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.407
Seven defects reported by MIF.
- anchors test selects a stale test that only a hand-ingested report timed. A test file belongs to the suite whose report timed it. A time recorded under a report no declared suite writes (anchors ingest --junit out/fx.xml) made the file belong to no suite. Every selection then left it out, and did not count it either: anchors test read "318 fresh and passing" while the commit hook refused four stale proofs. Such a time now names no suite, and the file belongs to the suite of the layer that ran it. (SLCTN-B20)
- A file given where a layer goes points at --changed. In anchors test unit -w backend a.test.ts, the path was read as an undeclared layer. The same happened to the second file of --changed a.ts b.ts. The error now says which arguments are files and prints the command that runs them, with one --changed per file. The flag's help says it takes one value per flag, or values separated by commas. (STPRS-E10)
- anchors audit <file> leaves out the gates of the whole project and the slow ones. Every audit ran the SBOM over the repository, the typecheck and the dead-code search: 10 to 30 minutes per file. The dossier now skips the gates declared scope: project or cost: slow, names them, and points at anchors check --changed <file>, which still runs them. (DTAUI-B10)
- A test suite can exclude: files its paths: cover. Two suites of one workspace often share a root and are told apart by the runner, as jest --selectProjects unit does. Anchors cannot read that selection. A new integration test was handed to the unit suite, which ran none of it, so every run failed with "reported a result for none". exclude: takes globs and is validated like paths:. (CNFGO-B79)
tests:
- workspace: backend
layer: unit
paths: ["packages/backend/**"]
exclude: ["packages/backend/__tests__/integration/**"]
- states-required registers a state where the spec defines it, by its name. A state counts when it is defined in one of these places:
- a heading that opens with the state's code (the whole heading counts);
- a table row whose first cell is the code (that cell and the next count);
- a list item that opens with the code (up to the first spaced dash or period).
Before, any line that cited a state's code counted, with whatever word it carried. A message row "Empty state — X-S02" registered "empty" for a screen with no empty state of its own. Specs that relied on such a mention now fail and name the missing state: give it a heading, or waive it with @no-state: <name> — <reason>. (STRQS-B03)
- rule-uses-implemented honours @no-uses. A rule whose line carries @no-uses: <why> is not asked what it reads, as rule-uses-declared already did. (RLUSG-B08)
- env-declared accepts a declared variable that the unit's dependencies read. A spec may declare a variable its own code does not read. The variable now counts as read when a code file the unit depends on reads it. The gate follows depends-on from the specified files, transitively. A shared helper's read is part of the unit's deploy contract. Variables read only by a dependency are still not asked of the spec. (ENVDC-B07)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- ad31a56 fix: seven defects from MIF — stale tests timed by hand, audit cost, suite exclude, gate precision
0.1.406 2026年11月12日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.406
- A merge commit counts as the last commit of the files it changed. The updated-at-atual gate and the map read each file's last commit date from one git log --name-only, and a merge lists no files there. A file the pre-commit dated in a merge then read the date of the commit before it: on the next commit the gate charged updated_at (2026-11-12) ≠ last commit (2026-11-11), and no edit to the file could satisfy it. The log now lists, for a merge, the files it changed against every parent, as git log -- <file> already counted them: a header dated in the merge, a conflict resolved. A file that came unchanged from one side keeps the day of that side's commit, so a merge made without the hook (a hosting service's merge button) still passes. Reported by MIF. (GTMTG-B18)
The first anchors map build after the upgrade may change the date of the nodes a past merge changed.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- c329cfc fix(gitmeta): a merge commit is the last commit of the files it changed
0.1.405 2026年11月12日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.405
- New gate symbol-exists: a name marked as a symbol is in the project's code. A comment or a spec marks the names it cites with @@name, #name or $name (dotted parts too: #Reader.open). The gate reads the marks in a code file's comments and in a spec's text outside its fenced blocks. Each mark is confronted against the identifiers of the map's code and test files, with their comments removed. A name that is not found fails, naming file:line and the name. A line that cites a name on purpose (one a dependency removed) carries @no-symbol-check: <reason>. Only marked names are read, so an existing project adopts the gate by marking its comments and specs; nothing unmarked is confronted. A mark glued to a word, a path or a link (a#b, /x#y, ](#section)), a name followed by a dash (#css-id), an editor's #region/#endregion, a color (#FFF, #C9BCFF) and the # that opens a comment are not marks. It is in the catalog, informative; declare it with blocking: true to make it block. From MIF: a comment explained a rule by isEvalSupported, an option pdfjs-dist 5 removed, with every gate green. (SYMXS)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- ce54d24 feat(gate): symbol-exists — a name marked as a symbol is in the project's code
0.1.404 2026年11月12日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.404
- In a git worktree, the pre-commit defers to the commit-msg as in the main tree. The pre-commit looked for the commit-msg under $(git rev-parse --git-dir)/hooks. In a worktree that is .git/worktrees/<name>, which has no hooks/. The pre-commit concluded there was no commit-msg and blocked at once, so a [skip-<gate>@<CODE>: why] in the message was never read. It now looks where git runs hooks from (git rev-parse --git-path hooks, which follows worktrees and core.hooksPath), and so does pre-commit.d. Reported by MIF. (INHKN-B19)
Run anchors install-hooks --force to get the new pre-commit.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 08373a5 fix(hooks): in a worktree the pre-commit finds the commit-msg where git runs hooks from
0.1.403 2026年11月11日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.403
Three fixes reported by MIF. The first two are in anchors test: a run that proved nothing no longer reads as success.
- The project root is written as the disk names it. On macOS the shell's working directory may differ in case from the folder (.../mif for .../MIF). Anchors built every path from it, so jest --findRelatedTests got paths it did not know, found no test and exited 0, in five suites. The root now takes the disk's case, read on macOS from the kernel and on Windows from each folder's listing; a part under a symbolic link keeps the link as written. (PRRPR-B04)
- A run that proved none of its files fails. When a suite reports a result for none of the files handed to it, the run fails, naming the count and the first path, whatever the suite's exit code. When only some files come back without a result, they are listed as unproven, as in v0.1.401. (SLCTN-B19)
- A @contract stamp holds across a change only to comments. mock-stamped hashed the stamped block with its comments, so a rule's mark edited in a comment broke the stamp. In MIF, // RPBTR-B02: → // RPBTR-B02/RPBTR-A01: broke 81 stamps, and 122 in all, each asking a stamp --refresh that reconfirms nothing. When the hash no longer matches, the gate now finds the version the stamp was taken at in the module's recent history. If the block's code is the same today, with comments out and blank lines left out, the stamp holds. The comment can be edited, added at a line's end, or added on a line of its own. A change to the code still fails, as before. Stamps written before this version need nothing. Reported by MIF. (MCSTM-B24)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- f369cc8 fix(config): the root takes the disk's case on Windows too
- 762ebff fix(config): the root's link-free parts are compared ignoring case, as Windows resolves both
- eaa58b8 fix(gate): a @contract stamp holds across a change only to comments
- 1c545fe fix(test): the root takes the disk's case, and a run that proved none of its files fails
0.1.402 2026年11月11日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.402
- push_review.require_gates — a waiver in a commit message frees the commit, not the push. With it, the pre-push hook runs the blocking gates a commit or a push charges over what the pushed commits reach. Every file is read as the pushed tip holds it, and no [skip-<gate>@<CODE>: why] from a commit message counts. Each failure refuses the push naming the gate, the target and the pushed commit whose message waived it. An edit not yet committed is not what the push sends, and does not count. A waiver meant to reach the remote goes in the artifact (@no-*), which the gate reads. It works with or without required and require_judgment. Requested by MIF. (PRCMS-B09, PSRSA-B01, RPSCR-B62)
push_review:
require_gates: true
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 0367c0e feat(review): push_review.require_gates — a commit-message waiver frees the commit, not the push
0.1.401 2026年11月11日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.401
Two fixes in anchors test, both reported by MIF.
- A stale test runs by default, whatever its last result. The default selection left out stale test files whose last result had passed ("stale but passing", behind --include-passing). A stale result is no proof, and evidence-fresh, which blocks, refuses it. So the default run read green while the gate stayed red: MIF reached evidence-fresh ✗578 with anchors test selecting 37 files. Every stale test file now runs by default. --include-passing now applies to mutation runs, where a stale score at the floor still waits for it. (SLCTN-B17)
- A file handed to a suite that comes back with no result is named. A suite whose paths: cover files its runner does not run (Maestro flows under a jest suite's apps/mobile/**) was handed them, proved nothing, and the run read as if they had run. The run now lists them as unproven and says how to fix it: narrow the suite's paths: to what its runner runs, and declare the suite that runs the others. (SLCTN-B18)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 59152e8 fix(test): a stale test runs by default, and a file the suite does not run is named
0.1.400 2026年11月11日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.400
- push_review: — a push carries its review, and the judgments its commits reach. Opt-in in anchors.yaml:
push_review:
required: true # each pushed commit is covered by a review whose verdict is ok
file: REVIEWS.md # where the reviews are written (the default)
retention: 30d # writing a review drops those older; empty keeps them all
require_judgment: true # each judgment gate the pushed commits reach has a verdict in the pushed map
Whoever pushes reviews what the push sends and records it: anchors push-review --ok "<what was validated, and how>" --by <who>, or --fail "<what failed>", with --note per observation. The review goes to the top of the file (when, who, verdict, commits with their patch ids, the judgment gates' answers, the text) and is committed alone. The pre-push hook runs anchors push-review verify with the refs the push sends. It refuses:
- a commit no review covers;
- a commit whose newest review failed;
- with require_judgment, a commit reaching a judgment gate that has no verdict in the map the push sends.
A review covers a commit by its hash, or by its patch id after a rebase. A commit that changes only the reviews' file and the map owes no review. anchors guide review has the new section "Before the push". (PSRSA, PRVLS, PRCMS, INHKN-B18, GVGDG-B32)
Run anchors install-hooks --force to get the new pre-push; without push_review: it confronts nothing.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 35191b5 feat(review): push_review — a push carries its review and the judgments its commits reach
- 03ae5ad test(review): the push-review tests give git an identity, as a CI runner has none
0.1.399 2026年11月11日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.399
- A block comment inside a line of code no longer stales the file's proofs. The comment was blanked in place, so load(a /* ARSCA-S04 */) read as load(a followed by spaces and ), the evidence revision moved, and every proof of the file had to run again. It now leaves the line as if never written: it goes out with the blank before it when a blank, a closer or the end of the line follows, and with the blank after it when an opener precedes it. A whole-line comment and a comment at a line's end were already left out. Reported by MIF while marking rules in code. (RPSCR-B61)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- dd5f234 fix(scan): a block comment inside a line of code leaves the evidence as if never written
0.1.398 2026年11月11日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.398
- The spec guide names two more cases as variants: a screen that reads a status the backend writes, one variant per value (@variants: active, past_due, suspended), and a write behind an authorisation, which also refuses (@variants: allowed, denied), the denied scenario proving the refusal where the write happens.
- The anchors.yaml page shows boundaries: as it is read — layer:, a forbid: regular expression and because: — instead of a from:/list form the configuration never accepted, with a second example: a backend enum shown raw on a screen (<Text>{item.kind}</Text>) forbidden in the screen layer.
These close DESIGN-proof-gaps: of MIF's proposals, the rest is either covered by what exists (nav-symmetric for a promise between two screens), the project's own (eslint rules, typed label maps, TZ in its test runs), or left open until a peer shows a case the current gates miss.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 56997e0 docs(guide): status and authorisation as variants; boundaries example as the config reads it
0.1.397 2026年11月11日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.397
- required_states: and the gate states-required: every spec of a layer registers the states the project requires of it. A spec registers the states its author thought of, and every state gate confronts only those; MIF traced nine released defects to a screen state nobody wrote (an error shown as an empty list, a not-found that spun forever). The project names the states each layer owes, by the layer in the spec's header, and each spec of it registers a state by each name — a line carrying the state's code and the name — or waives it with @no-state: <name> — <reason>. The names are the project's; Anchors proposes none. Informative, in the catalog; asks nothing until required_states: is declared. (CNFGO-B78, STRQS)
required_states:
screen: [loading, error, empty, offline]
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 49d22f1 feat(gate): required_states — every spec of a layer registers the states the project requires
0.1.396 2026年11月11日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.396
- anchors map deps <unit> --flat: a unit's dependency closure as paths, one per line. The tree map deps prints, down or with --up up, within --depth, as sorted paths with the file itself included — the shape a project's script reads to cross a handler's closure with the environment it reads, the tables it touches or the grants it needs. From MIF's P12: five released defects were wiring (an env var, a table or a secret not granted to the handler that reached it through a dependency). (MDCMP-B04)
- test_bypass: (v0.1.395) is documented on the anchors.yaml page and in anchors guide test.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 4f092e1 feat(map): map deps --flat prints a unit's dependency closure as sorted paths
0.1.395 2026年11月11日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.395
- test_bypass: and the gate bypass-proven: a path the code skips while a test runs is run by a suite that reaches it. From MIF-1S: a hook returned early under EXPO_PUBLIC_E2E, the unit tests doubled the write, the end-to-end skipped it, and nothing ran it. The project declares how a bypass is recognised and which suite layers run the skipped path; a code file with a bypass asks a test of its own that passed in one of them, or @no-bypass-proof: <reason> on the bypass's line or the line above. Informative, in the catalog; asks nothing until test_bypass: is declared. (TBYSA, BYPRS)
test_bypass:
patterns:
- "process\\.env\\.EXPO_PUBLIC_E2E === 'true'\\)\\s*return"
proven_by: [integration]
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 193ba4e feat(gate): test_bypass — a path skipped while a test runs is run by a suite that reaches it
0.1.394 2026年11月11日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.394
- external_systems: — the real systems the code talks to, and the suites that reach them. A unit test proves a file against doubles, and a double of a database or an API accepts whatever it is given. MIF traced fourteen released defects to that (an index that did not exist, a condition outside the key schema, a field the API refused) and nine more to recorded responses of an API version that was gone. (EXSSA)
external_systems:
- name: appsync
calls: "\\bclient\\.graphql\\("
proven_by: [integration, live]
- name: stripe
calls: "\\bstripe\\.[a-z]+\\.(create|update)\\("
proven_by: [live]
version: 2025-09-30.dahlia
max_age: 90d
The block is not boundaries:, which keeps the layers' import boundaries.
- New gate external-proven: a code file that calls a declared system needs a test of its own — derived beside it, importing it, or testing its spec's feature — that passed in one of the system's proven_by suite layers. @no-external-proof: <reason> waives it for a file. (EXPRS)
- New gate recording-current: a recorded response says where it came from, @recorded: stripe 2025-09-30.dahlia 2026-10-02. One of another version than the system's version, or of an undeclared system, fails; one older than max_age diverges until it is recorded again. (RCCRS)
Both gates are informative in the catalog and ask nothing until external_systems: is declared. The tests guide explains the two.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- f983401 feat(gate): external_systems — a file calling a real system needs a proof that reached it
0.1.392 2026年11月10日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.392
- A persisted data ties its writer to its reader: writes: and reads: in the header. The writer and the reader of one record often never import each other, so nothing tied them, and each side stayed green against its own fixture while they disagreed on a field, a sign or a vocabulary — seven of the defects MIF traced to releases. A header now says what the file persists and consumes, with names the project chooses:
// @anchors
// code: ALRTS
// writes: NotificationLog
// reads: AuditItem, NotificationLog
The map ties each reader to every writer of its data (reads-from, GRBLG-B33), and a reader's test evidence holds the writer — not what the writer itself uses —, so a change to the writer's shape asks the reader's tests again through evidence-fresh. (RPSCR-B60, EVFRA-B16)
- New gate reads-written: a data a header reads that no header writes ties nothing, and the gate names it — a typo, or a writer that never declared itself. Informative, in the catalog; it asks nothing of a file whose header reads nothing. (RDWRS)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 2b5fed1 feat(map): writes:/reads: tie a data's reader to its writer, and the writer stales the proof
0.1.391 2026年11月10日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.391
- New judgment gate rule-test-coherent: do the tests assert what the rule says? The pair of rule-fulfilled on the test's side. For each rule of a spec, the judge reads the rule's text and what the tests carrying its code assert — the targets of their expect/assert, the functions, tables and fields they check — and fails a contradiction, naming the rule, the test and both sides. From MIF's MIF-1S: the rule said a tap records the item in the notification log, the test with its tag asserted markAuditSeen, and the tag and a green test were all any gate saw. (DFGTD-B22)
A new project is seeded with it. An existing one opts in with one line in gates::
- name: rule-test-coherent
- The spec guide names the cases a deterministic test never meets unless asked, as variants (variants-proven, v0.1.390): a matching or timing rule's refusal (@variants: matches, does-not-match), order (a-first, b-first), idempotency (once, repeated), concurrency (alone, interleaved), and a timer stopped on every exit (completes, leaves). Each then asks its own scenario.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 1bddcc0 feat(gate): rule-test-coherent — judge whether the tests assert what the rule says
0.1.390 2026年11月10日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.390
- New gate variants-proven: a rule that names its variants is proven once per variant. A rule that holds for several values of a discriminant — an item that is an audit or a notification — names them on its line, @variants: audit, notification. Each variant then asks a scenario of the rule that names it: a @variant:<value> tag beside the rule's tag, or a row of an outline's Examples. A variant with no scenario fails, naming the rule and the variant. (VRPRS-B01)
| `UAUDU-B02` | a tap marks the item seen @variants: audit, notification |
| `UAUDU-E02` | a failed write propagates @variants: B02 |
@UAUDU-B02 @variant:notification
Scenario: tapping a notification records it in the notification log
- A rule may follow another's variants: @variants: B02. An error or an effect shared by every kind asks each of the variants of the rule it follows, so a new value asks a scenario of every follower. While one is missing, the gate diverges: the follower is "proven only without" it. (VRPRS-B02)
The gate is in the catalog as informative, and asks nothing until a rule writes @variants:. The spec and feature guides explain the marker. From MIF's case MIF-1S: one scenario proved the audit, the notification path shipped broken, and every gate — mutation at 100% — was green.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- c6e3abd feat(gate): variants-proven — a rule naming its variants asks a scenario per variant
0.1.389 2026年11月10日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.389
- A gate's legacy can end: legacy.until. legacy: softens a gate for the targets last changed before a date, and until now it did so forever. until: <date> is its deadline: from that day on no target is legacy, and every one answers with the gate's own severity. Before it, a verdict the legacy softened says until when, and what it will do after. A new gate can now bar the new code at once and the old code from a date on, instead of leaving the old code pending without end. (GTENG-B37, CNFGO-B77, CNFGO-E13)
- name: mutation-score
blocking: true
legacy:
before: 2026-10-21
until: 2026-12-15
severity: {fail: inform}
An until that is no date, or does not come after before, refuses the configuration.
This is the first item of DESIGN-proof-gaps.md, the plan built from the defects MIF traced to releases with every gate green; each item ships opt-in.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 386c687 feat(gate): legacy.until — a gate's legacy ends on a deadline, and the old code answers as the new
0.1.388 2026年11月10日 — 64ビット ZIPARM64 ZIP
リリースノート
v0.1.388
Three fixes reported by SS while rewriting its app specs.
- @no-mark without a target, on a rule's line, waives only that rule. It used to waive every rule of the spec: one @no-mark: in a single table row silenced rule-implemented for the whole file. Written on a line that declares no rule (a paragraph, a quote), it still waives the whole spec, as before. A spec relying on a row's waiver to cover the others will now see them named. (RLIMR-B06)
- check --fix no longer gives two files the same code in one run. The codes were picked against the map read before the run, so files fixed in the same batch could each find the same code free (12 collisions over 69 headers). (FXIXX-B10)
- A commit-message waiver [skip-<gate>@<CODE>: why] accepts the file's own code: as well as its unit's (the spec's code, the prefix of its rules and scenarios), and anchors guide gates <name> says which codes it takes. (GTENG-B09)
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 3cfd84f fix(gate): @no-mark on a rule's row waives that rule only; --fix gives no code twice
0.1.386 2026年11月7日 — 64ビット ZIPARM64 ZIP
リリースノート
New: aliases: — short names for the paths of anchors.yaml
A path written in anchors.yaml is often long, and the same prefix repeats across layers, guides,
derived templates and gate exceptions. aliases: names it once:
aliases:
"@app": apps/mobile/src
"@guides": docs/engineering/guides
layers:
ui:
pattern: "@app/screens/**/*.tsx"
kind: code
governs:
- from: "@guides/SCREEN_GUIDE.md"
governs: ui
An alias is named @ and a word and stands for a path relative to the root. It is translated as
the first segment of a path, in every field that holds one: layers, guides, derived,
frozen_paths, suites and their reports, docs, changelog, pipeline.workflows, gate
exceptions, dialect globs. The translation happens once, when the configuration loads; every
command reads the translated path.
A path that cites an alias aliases: does not declare refuses the configuration, naming the
field; a real folder named with @ (@scope/pkg) is kept as written. An alias that opens with
another, or names no path, is refused too.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- a91a7e0 feat(config): aliases — short names for the paths of anchors.yaml, translated in one place
0.1.385 2026年11月6日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: a stage's last job is anchors pipeline record
shellcheck, run on every workflow by actionlint, read anchors pipeline done <stage> as the end
of a shell loop and reported SC1010 on each workflow that calls it (MIF).
The command that runs a stage's verify: and records its outcome is now anchors pipeline record <stage>. done stays as an alias, so a workflow already written with it still runs. The
pipeline-wired gate reads record, done and run as a stage's record.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 5ff184c fix(pipeline): the stage's last job is pipeline record, with done as its alias
0.1.384 2026年11月6日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: passed: reads the stage's record on the sha itself
passed: <stage> looked for the stage's commit status only on the head of the pull request that
made the sha. A later stage of the same branch could not see a stage recorded on that same sha.
At MIF, the version pointer is meant to flip only after the production release passed on the
production sha, but it looked for that release on the review commit, which never carries it.
passed: now holds when the sha itself carries the stage's success. Failing that, it holds when
the head the sha was merged from carries it, as before.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- bca5e5b fix(pipeline): passed: reads the stage's record on the sha itself, then the merged head
0.1.383 2026年11月6日 — 64ビット ZIPARM64 ZIP
リリースノート
Feature: pipeline stages ask the Anchors before they run
A release stage ran on a sha no review had seen, because nothing the workflow consulted said it
should not (MIF). pipeline: declares the stages a pipeline asks about: the branches each runs
on, the conditions its sha must meet, and the command that validates it.
pipeline:
stages:
review-release:
branches: [review]
requires: [flow, { gates: ci }]
verify: node scripts/smoke-review.mjs
production-release:
branches: [production]
requires:
- flow
- any: [{ passed: review-release }, { from: "hotfix/*" }]
verify: node scripts/smoke-prod.mjs
The three commands, through the gh CLI with GH_TOKEN:
- anchors pipeline can-run <stage> is the first job; the stage's other jobs need it. It refuses
a stage that is not declared, a branch the stage does not run on, and a requirement that does
not hold, naming each.
- anchors pipeline done <stage> is the last job. It runs verify: and records the outcome on
the sha as the commit status anchors/pipeline/<stage>.
- anchors pipeline run <stage> does can-run, run:, verify: and the record in one step.
The requirements:
- flow: the sha is the merge of a pull request branch_flow: allows.
- gates: <phase>: the phase's gates pass.
- passed: <stage>: the head the sha was merged from carries that stage's success.
- from: <branch>: the sha was merged from that head.
- any:: one of several holds.
The pipeline-wired gate, declared with scope: project, reads .github/workflows/*.yml. It
fails on four things:
- a call to an undeclared stage;
- a declared stage no workflow calls;
- a stage asked and never recorded;
- a workflow that calls no stage and is not under pipeline.unguarded.
A project with no pipeline: sees no change.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 355e14d feat(pipeline): stages ask before they run, record when they pass; pipeline-wired reads workflows
0.1.382 2026年11月6日 — 64ビット ZIPARM64 ZIP
リリースノート
Feature: the branch flow — which branch merges into which
A release flow goes one way, and nothing said which. At MIF, commits went on the production
branch and the review branch was aligned from it, the flow run backwards, and nothing noticed.
branch_flow: declares it: for each base branch, the heads a pull request may merge into it,
and the branches that take no commit of their own.
branch_flow:
merges:
review: [develop, "hotfix/*"]
production: [review, "hotfix/*"]
develop: [review, production, "hotfix/*"]
protected: [review, production]
A branch is a glob whose * does not cross /, or a regular expression after re:.
- anchors branch check, on the pull request event, refuses a pair the flow does not allow.
- On the push event, it refuses a push to a guarded branch that is not the merge of an allowed
pull request. It asks GitHub through gh with GH_TOKEN. This covers a repository with no
server-side protection, such as a private one on GitHub's free plan.
- On the machine, the pre-commit and a new pre-merge-commit hook refuse a commit on a protected
branch (anchors branch guard). Run anchors install-hooks --force to get the new hook.
A project with no branch_flow: sees no change.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 7d46c90 feat(branch): the branch flow — which branch merges into which, and which take no commit
0.1.381 2026年11月6日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: an operation's security schemes are required together
An API spec's ## Security table lists the schemes an operation requires. Each row compiled to an
item of its own under security:, and OpenAPI reads each item as an ALTERNATIVE. A client or a
gateway reading the document then accepted a call carrying any one of the schemes. At SS, a
session and an App Check, both required, compiled as either one.
The rows of a Security table now compile to one requirement holding every scheme, each with its
scopes:
security:
- firebaseSession: []
appCheck: []
A table with one row compiles as before.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 64fe80f fix(openapi): the rows of a Security table are one requirement, every scheme together
0.1.380 2026年11月5日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: a spec's pieces are its unit's layer's, in every command
A spec names its unit's layer in its header (layer: api-handler), and the map derives the
unit's pieces by that layer: a derived.overrides entry with when: api-handler gives
unit-complete its <op>_test.go. The commands that derive the same pieces from a path used the
spec's own layer, spec, so they never read the override. At SS, anchors work code --for <op>.spec.md prescribed <op>.test.ts for a Go handler.
work, check --changed, deliver and watch now derive a unit's pieces by the layer the
spec's header declares, as the map does. work names that layer next to the file's own
(Unit layer: **api-handler**).
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 34ae9ed fix(derived): a spec's pieces are its header layer's in work, check, deliver and watch
0.1.379 2026年11月5日 — 64ビット ZIPARM64 ZIP
リリースノート
Feature: an OpenAPI field can be another contract
A contract's field holding another contract had no type: its Type cell was kept as the field's
description, and the clients generated from the document typed it as anything. At SS, 19 fields
came out untyped, the items of 17 pages among them.
A Type cell now cites a contract by its code in backticks, as the body and the responses do:
─────┬─────────────
Input│Type
─────┼─────────────
items│array of
│CBRNC
─────┼─────────────
rule │RCRUL
─────┴─────────────
- `CODE` is a $ref to that contract's schema; array of `CODE` and `CODE`[] are
an array whose items is that $ref. The field keeps what it accepts as its description.
- The cited contract is a shared schema in components, even when only a field cites it.
- A contract citing itself, or one citing it back, compiles: each is a reference.
- A code no spec of the project has fails the build, naming the spec and the code (OPNAP-E03).
A Type cell with no code in backticks reads as before.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 8237e4f feat(openapi): a field typed with a contract's code is a $ref to its schema
0.1.378 2026年11月3日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: a name with accents gives a code of letters
anchors code and anchors new compress a unit's name into its code letter by letter, and read
the name byte by byte: "Resposta à proposta" gave R followed by half of the à, a code that is no
code (reported from SS, which writes its specs in Portuguese).
A name's letters are now taken without their accents (à → a, ç → c), and any other character
outside A–Z and 0–9 ends a word. "Resposta à proposta" gives the code of "Resposta a proposta". Codes
already in headers do not change: only new ones are generated.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 2009bd3 fix(code): a name with accents gives a code of letters
0.1.377 2026年11月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: what anchors init seeds passes its own gates
Two defects found starting the SS monorepo:
- In a project of several languages, the configuration init wrote did not load. Each family's
example carries the naming conventions (set_promise, set_slice), and init wrote them inside
every dialect of dialects:, where they are the project's: the next command failed with
dialects.go.set_promise: the project's, not a dialect's. They now go in dialect:.
- The header guide failed the header gate. guides/HEADER_GUIDE.md — and the spec guide — were
seeded with no @anchors header, and a project with a guide layer had its first commit barred
by header-valid on the very guide that teaches the header. Each seeded guide now opens with its
header, under a code no file of the project holds.
And the doctor's guide-sem-governo on a fresh project: a seeded guide the configuration reads as a
guide now governs the layers it rules — the header guide the layers whose files carry a header, the
spec guide the specs.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 22bf5d6 fix(init): what init seeds passes its own gates
0.1.376 2026年11月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: a new coverage measurement replaces the old ones
A file's coverage keeps one entry per suite, and the node is as fresh as its stalest entry. A
suite measured before the file changed stayed beside a fresh measurement, and the node read stale
until that very suite ran again. At MIF, one report written a second before its file's last edit
kept a handler stale for good.
A coverage ingestion of the file's current content now removes every entry measured against
another revision, or with none: a new measurement always replaces the old ones. A report older
than the file is itself the old measurement: beside a fresh one it is dropped, and alone it is
kept with no revision, as before. With nothing measured since a change, the node stays stale.
Feature: a file git moved keeps its evidence
A node of the map is known by its path, so a git mv lost every proof of a file whose content had
not changed. At MIF, 96 files moved into a package each asked for their mutation again.
A rebuild now reads the moves git sees since HEAD (git diff -M, staged or in the tree) and moves
each node to its new path before carrying the previous map, with the edges at it and what tests
recorded of it. The content then decides, as for any edit: the same content keeps everything, a
change that proves nothing (a header date, a comment) carries it, and a change to the code drops
it. map build, the commit's map, docs build and recode do this, and keep-evidence finds a
moved file in HEAD's map at its old path.
git decides what a move is, by content: a small file moved and much edited is a deletion and a new
file, and starts with no evidence.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 3231324 fix(coverage): a new measurement replaces the old; feat(map): a file git moved keeps its evidence
0.1.375 2026年11月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Feature: a dialect per layer — a project in several languages
dialect: was one block per project, so a project could speak one language. A repository with
TypeScript apps and Go services had its Go code read with the TS lexicon: the patterns matched
nothing, and the gates passed it green.
A project now declares each language's lexicon once, by name, and each layer names the one its
files are written in (DESIGN-dialect-per-layer.md):
dialect:
gherkin_language: en
dialects:
ts: { family: ts, ... }
go: { family: go, ... }
layers:
screen: { pattern: "apps/**/*.tsx", kind: code, dialect: ts }
handler: { pattern: "services/**/*.go", kind: code, dialect: go }
go-test: { pattern: "services/**/*_test.go", kind: test, dialect: go }
- A file is read with its layer's dialect alone: its comments, imports, exports, tests, the
failures it handles, the statuses it emits, the variables it reads. A file already belongs to
one layer, and priority settles two matching patterns.
- Beside dialects:, dialect: holds only gherkin_language, set_promise and set_slice.
- With several dialects, every code and test layer names one, or the configuration does not
load. Spec, feature and doc layers name none.
- Who imports a file is read from each importer with its own dialect. The test handle patterns
are every dialect's together: a flow queries what a screen of another language exposes.
- A spec whose code is written in two dialects leaves the failure gates pending, naming them.
- A gate presupposing a field is pending at each dialect missing it, named there
(dialects.go.env_read).
- anchors doctor names a layer holding files of a family its dialect's is not, with that
family's example; its other tips speak of each dialect at its place.
- anchors init proposes one dialect per family it finds in the code roots, and a test layer per
family.
A project of one language keeps its dialect: block, and nothing changes for it.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 76a8b3e feat(dialect): a dialect per layer — a project in several languages
0.1.374 2026年10月25日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: evidence kept by declaration survives the next change that proves nothing
anchors keep-evidence moved a file's evidence to its current content and moved the node's
revision with it, but left the node's evidence revisions — the content without what proves
nothing, such as the header's date — those of the previous content. The next rebuild compared
the new content's evidence revisions with those, found them changed, and dropped what the
declaration had kept. At MIF, keep-evidence --lines on 781 files whose imports alone had
changed, followed by anchors touch bumping their headers' dates, left 271 of them with no
mutation.
The node now takes the evidence revisions of the content kept. The same holds for the files
check --fix repairs and for the specs anchors migrate rewrites.
keep-evidence also brings back the mutation the map at HEAD holds for a file whose signal
holds none — a coverage run at the new content had started a new signal, and the command
found nothing to carry. A mutation measured again at the new content stays its own. To recover
files that lost their mutation this way, run keep-evidence --lines on them again before the
map is committed: the map at HEAD is where the mutation comes back from.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- a9c9092 fix(keep-evidence): the kept evidence survives a date; a mutation HEAD holds comes back
0.1.373 2026年10月24日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: a package's double is not the project's file of the same name
Reported from MIF: jest.mock('stripe', …) in infra/stripeBalance.test.ts doubles the npm
package stripe, and anchors stamp stamped it against the project's own infra/stripe.ts.
A double's specifier the import resolution does not reach was matched by its tail, and a bare
name of one segment matched any file of that basename in any folder. The test was tied to a
file it never reads, and every edit there would have charged a divergent stamp.
A bare name of one segment that no alias maps (stripe, uuid) is now the project's only as
a file at the root. Everywhere else it is a package's: mock-typed and mock-stamped do not
charge it, anchors stamp does not stamp it, and the map no longer records the test as
doubling the file. A path with folders (src/a), a relative path and an alias resolve as
before.
At MIF's HEAD one more double changes: jest.mock('uuid') in apps/mobile/src/utils/uuid.test.ts
no longer names apps/mobile/src/utils/uuid.ts, so that test's proof reaches the file it tests.
The @contract: stamps written against uuid.ts and infra/stripe.ts for those packages are
no longer asked for, and can be removed.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 9eed79f fix(mock): a package's double is not the project's file that shares its name
0.1.372 2026年10月24日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: pagination-honored reads each function's own text
Reported from MIF (packages/backend/models/adminSources.ts): two sister functions read through
a local readAll that walks every page, each with // @no-paginate: <reason> above its
signature. The first passed and the second stayed accused; only a second mark inside its body
made it pass. Three things were wrong, and all three are fixed:
- The mark above a signature is that function's. A function's body ran to the next
declaration, so the mark above the second function was read as the first one's. The comments
right above a declaration now belong to it, wherever it stands in the module.
- A helper of the module that drains the cursor paginates its callers. A function that
reads through a function the module defines (dialect.definition) whose code walks the
cursor in a loop now passes with no mark. The mark stays for what reads one page on purpose.
- A comment is no code. The gate read loops and cursors in comments too: the Portuguese
do of "o uso do mês" counted as a loop, and functions passed as paginating by accident.
Comments are now blanked before the code is read; the mark, a comment, is still read.
At MIF's HEAD the verdicts of the 109 files with a collection query are unchanged; the marks in
adminSources.ts can now be removed.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 3fafddb ci(release): the Winget submission gets a second attempt a minute later
- 1811254 fix(pagination): a waiver above a signature is its function's; a draining helper paginates
0.1.371 2026年10月24日 — 64ビット ZIPARM64 ZIP
リリースノート
Docs in the project's language, orphan pages removed, and a calmer progress view
The progress shows detail only where something is moving. A settled demand is one that
is all done, or that has none of its tasks started and none in progress.
- progress summary and the monitor print a settled demand as one line, without its
stages.
- progress show, for the whole file, prints a settled demand as its heading only,
without its stages or tasks.
- What keeps everything: progress show <demand> still details that demand, and the
file and --json still keep everything.
docs build removes the pages its templates no longer produce. A page in docs/ whose
generated marker names a template of doct/ that is gone is removed, along with any folder
left empty, and the build lists each removal. This covers renamed templates and deleted
ones.
- A page without the marker is never touched.
- A page listed in docs.keep is never touched either. The list takes globs relative to
docs/, for pages generated once and kept by hand since:
docs:
keep: ["guides/**"]
- --dry-run lists what it would remove and removes nothing.
The layer page's summary note follows the project's language. It used to be in
Portuguese whatever lang: said.
This repository's docs are now in English: architecture.md, behavior.md, rules.md
and layers/gate.md. docs init already scaffolds in the project's language. A project
whose templates predate that can delete them and run docs init again, and the next build
removes the old pages.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 459e6fb feat: docs in the project's language, orphan pages removed, settled demands as one line
0.1.369 2026年10月21日 — 64ビット ZIPARM64 ZIP
リリースノート
A cutoff date for legacy code, a result cache for project tools, and a truthful warning
legacy lets a gate be stricter on new code than on old code. A project that tightens a
gate can now leave the code written before a date under the old rule:
- name: mutation-score
severity: {pending: block}
legacy:
before: 2026-10-21
severity: {pending: inform}
- A target whose last commit is older than before, and that has no uncommitted change, uses
legacy.severity.
- A target changed since then uses the gate's own severity: one committed after the cutoff,
one with uncommitted changes, or a new one. Whoever touches legacy code takes on the full
rule.
- Only what the verdict does changes. The legacy findings are still reported, they just do
not block.
- legacy.severity can never be stronger than the gate's own, and a level it leaves out
takes the gate's.
cache keeps a project tool's verdict over the same inputs. It is opt-in for each gate:
- name: eslint
run: "bash scripts/anchors-eslint.sh {{files}}"
cache:
inputs: ["**/eslint.config.*", "pnpm-lock.yaml", "scripts/anchors-eslint.sh"]
- The tool runs again only when something its verdict depends on changed:
- the gate's declaration;
- the content of each target;
- the content of each file the inputs match;
- the version of Anchors.
- A pass is kept. A failure is kept only with failures: true. In a CI the cache is used only
with ci: true.
- A tool that did not get to run is never kept.
- A replayed verdict says it came from the cache, and when it was measured.
- check --no-cache runs every tool.
- The cache lives in .anchors/cache/tools/, outside git.
The "nothing to measure" warning tells the truth in partial checks. In a check --changed, a gate whose requires matched nothing in the changed files was reported as if
nothing in the map matched it. Outside --all, the warning now names only a gate that no
node of the whole map reaches by its on: and tags:.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- b558382 feat(gate): legacy cutoff, opt-in tool result cache, truthful nothing-to-measure warning
- a096e47 test(gate): the tool cache test counts runs in the project, the same under every shell
0.1.368 2026年10月21日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: a handle of values alone is measured through its file's own head
A component can give its handles a head of its own, as the default of the prop that carries
it, and build each handle from values only:
function RolePicker({ testIDPrefix = ':rlpc-role' }) {
return roles.map((r) => <Pressable testID={`${testIDPrefix}-${r.key}`} />)
}
testid-consistent treated ${testIDPrefix}-${r.key} as a handle made of values alone, and
left the spec pending. The head is in the text, though: it is the default the exposed
patterns already read as a prefix (rlpc-role-*). When a file's patterns find a head of its
own, such a handle is now that head's and is measured through it. At MIF, the two pickers
that were pending (RolePicker, SeatTypePicker) now pass.
A handle of values alone in a file that gives no head of its own stays pending, as before.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 150f0e5 fix(gate): a handle of values alone is measured through its file's own head
0.1.367 2026年10月20日 — 64ビット ZIPARM64 ZIP
リリースノート
Gates in parallel, a faster map, and late reports stamped as their run found the files
The gates run side by side.
- The gates Anchors answers itself run on every core.
- The project's tools run beside them, a few at a time.
- A tool that writes something another tool reads declares exclusive: true on its gate,
and runs alone after the others.
- Results come back in the order of the gates, as before.
- anchors check --sequential runs one gate after another, to rule the parallel run out
when a result looks wrong.
A parallel run surfaced a race: a gate that loads the configuration rewrote shared values
while other gates read them. Those values are now written only when they change.
The map is faster.
- map build reads and parses the files on every core. On MIF it went from 28 s to 8.6 s,
and the map came out identical.
- check --all no longer walks the whole tree to compute a stale-map list it never prints.
A late report is stamped as its run found the files. A JUnit report ingested by hand
used to be stamped with the tree of the ingestion. A screen that changed after the run
then kept a fresh proof. Now:
- anchors ingest places the run in time:
- --ran-at, a time or the commit the run measured;
- otherwise the report's timestamp;
- otherwise the report file's modification time.
- Each file a proof stands on is compared with its content at the last commit before the
run, using the evidence the map already uses, where comments and the header do not count:
- the test;
- the files the test reaches;
- the specs whose scenarios the test proves.
- A file whose evidence changed since then is stamped changed-after-run, so the proof is
stale and names it.
- --alias OLD=NEW (repeatable) and --alias-file read a code renamed since the run. The
alias applies to the report's cases and paths, and to each file's content at the run's
commit:
- a unit's code renames all of that unit's scenarios;
- a scenario's code renames only that scenario.
A file whose only change is the rename keeps its proof fresh. The signal records ran_at
and remapped.
- A run that anchors test just made is ingested as before.
coverage-delta now tells an earlier 0% apart from no earlier measure. Before, both
were stored as an empty value, so a file whose previous coverage was 0% read as having no
baseline. Now:
- a file measured for the first time is skipped, because there is nothing to compare;
- an earlier measure of 0% is compared like any other, so a rise or a hold passes;
- a file never measured is still pending.
Coverage measured at 0% before this version starts being compared from its next
ingestion.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- a084a44 feat: parallel gates, a map scanned on every core, late reports stamped as their run found them
0.1.366 2026年10月20日 — 64ビット ZIPARM64 ZIP
リリースノート
Performance: check --all takes half the time on a large project
A check --all on a clone of MIF, with the slow external tools skipped, went from 681 s to
329 s. Every verdict stayed the same: the same counts in all 109 gates, and the same 2675
findings.
Three things were redone over and over:
- A gate's premises. Every run of a gate on a target checked whether the configuration
declares what the gate needs, by converting the whole configuration to YAML and back. This
was 28% of the run. It is now done once per gate in each run.
- The project files. Each gate re-read its target's file, so the same file was read about
50 times. Within a run, a file is now read once, and read again after each tool the project
runs, since a tool may write files.
- The shape of the map in evidence-fresh. The gate rebuilt the map's edges for every
test, three times. It now builds them once per map. Revisions are still read live, because
the watcher updates them in place. The gate went from 65 s to under 3 s.
What remains is mostly the project's own tools. At MIF, eslint alone takes 1m43s.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 611eb5c perf(check): a run asks each premise and reads each file once; evidence indexes the map once
0.1.365 2026年10月20日 — 64ビット ZIPARM64 ZIP
リリースノート
Freeze a folder, and leave a frozen part's work out
Freezing an area that is out of scope used to mean freezing it unit by unit. Its gates went
quiet, but its tasks were still handed out, the progress still counted it, and impact
still listed its files as work to redo.
A folder can be frozen. anchors freeze [CODE] --path '<glob>' writes the scope to
anchors.yaml:
frozen_paths:
- freeze: "PLNAP-F01" # the freeze's code, or the reason with no doctrine
paths:
- "apps/mobile/src/features/ir/**"
- The freeze's record stays in the doctrine. The globs only say where it applies.
- The map freezes every file a glob matches, as the file's own header mark would. A unit
whose spec falls in the folder is frozen whole, including tests that live elsewhere.
- A file's own mark wins over the folder's.
- freeze-consistent names a glob that matches no file.
- anchors thaw CODE removes the globs together with the marks. thaw --path '<glob>'
removes one glob, and freeze show lists them.
- The agent hook warns about a new file inside a frozen folder too.
The work of a frozen part is not handed out.
- anchors next leaves a task on a frozen file in the queue, waiting for the thaw. It claims
the next task instead, or, when there is none, says the queue is empty and counts the
tasks waiting.
- The cold start does not seed a frozen plan, and does not count a spec inside a frozen
folder as missing.
- In github mode, the claim runs in the pipeline. A card for a frozen unit is still handed
out there, with the warning.
anchors impact lists frozen parts apart, under their own heading with their freezes,
not among the parts to redo.
The progress leaves a frozen plan out. A demand whose plan is frozen, by its header or
by a frozen folder, leaves the totals of progress summary, the monitor and status, and
its row says it is frozen. The file's own lines are unchanged. A task names no unit, so to
leave out one task of a live plan, use progress cancel with a reason.
The freezes section follows the project's language: ## Congelamentos in a pt-BR
project. A document that already has the section under another language's title keeps it.
Run anchors map build after adding frozen_paths by hand.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 5ed9728 feat(freeze): a folder is frozen by glob, and a frozen part's work is left out
0.1.364 2026年10月20日 — 64ビット ZIPARM64 ZIP
リリースノート
Fixes: several --junit reports, one flag per module on a line, typeof import is a type
anchors ingest takes several --junit reports. Before, only the last one was ingested,
with no warning. Now each report is ingested in turn as a suite of its own, and the
--lcov and --mutation reports go with the first. A single --suite given with several
reports is refused: one name cannot stand for several suites.
One flag per module on a line. A line can hold two statements of the same module, as in
require('x') as typeof import('x'). The fixer wrote the flag twice
(// @dep: CODE // @dep: CODE). Now:
- the line takes one flag, which answers for both statements;
- dep-declared asks for that one flag, and dep-honored checks it against both statements;
- the flag is @dep[type]: only when every statement on the line is types only, so with the
require it is a value flag;
- a line that imports two different modules still takes a flag for each.
typeof import('x') is types only. The ts family example now reads
typeof import('…') as a statement of its own, ahead of the dynamic import('…') that took
the same path, and its type_import_pattern reads it as types only. The fixer flags it
@dep[type]:, so it no longer counts as a value edge in a test's evidence. A project that
copied the example into its anchors.yaml adds two things by hand:
- the new pattern to dialect.imports, before the require|import( one:
\btypeof\s+import\(\s*['"](?P<path>[^'"]+)['"]\s*\)(?:\.<ident>)?;
- the alternative ^\s*typeof\b| at the start of type_import_pattern.
Then run anchors check --fix.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- ca495a8 fix: several --junit reports, one flag per module on a line, typeof import is a type
0.1.363 2026年10月19日 — 64ビット ZIPARM64 ZIP
リリースノート
A test's proof stops at its doubles, and a proof tied to no code is pending
A test's evidence stops at the modules it doubles. A test that doubles a module, such as
jest.mock('../client'), never runs that module. Its proof expired anyway whenever that module
changed, or anything behind it, because the closure walked into the real code. The map now
records on each test the files its doubles replace: each specifier that derived.mock_detect
captures, resolved like an import to one file of the project. The closure stops at them,
wherever the walk meets them, not only at the test's own imports. A change to a doubled module
is caught by the double's stamp (mock-stamped). A specifier that names a third party's
package, or several files the test's folder cannot choose between, stops nothing.
A proof tied to no code is pending. evidence-fresh passed a test whose score was measured
against no code at all: its imports carried no @dep, so its closure was empty and no change
could expire it. The gate now answers pending when all of these hold:
- the project's dep-declared confronts tests (on: holds test);
- the test imports code of the project;
- its proof reaches none of that code.
The verdict names the imports, and anchors check --fix writes the flags. An import of types
only, or of a module the test doubles, does not count. A project that keeps dep-declared on
code alone sees no change.
Run anchors map build once to record the doubles.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 928d565 feat(evidence): a test's proof stops at its doubles; a proof tied to no code is pending
0.1.362 2026年10月19日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: a derived handle that ends in its segment can be queried
A handle built on the caller's head can end in its own segment, as in ${testId}-spin. It
was written *-spin-*, so a query only counted with a hyphen after the segment. b-spin
was never queried, and testid-consistent failed.
Such a handle is now *-spin: a consumer queries it when the segment comes after a head and
is followed by the end of the id or a hyphen, as in b-spin, ${id}-spin or
b-spin-overlay. b-spinner does not count. A caller's own handle is queried through it
the same way, as in :row-transfer-badge. A segment followed by a value, as in
${id}-row-${i}, stays *-row-*.
Declare a closed handle as written in Test IDs: *-spin.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- c1d2da9 fix(gate): a derived handle that ends in its segment is queried at the id's end
0.1.361 2026年10月19日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: a broken anchors.yaml is an error, never an absent one
The commands that read the project's configuration loaded it and ignored the error.
An anchors.yaml that does not load, for example one holding a field this binary does not
know, was treated as if there were no configuration. anchors touch then found no comment
syntax and said every file had no header. It dated nothing and still exited 0. Measured at
MIF, a chain left 16 headers with their old date.
These commands now fail with the load error:
- touch, verify, stamp --refresh, new, deliver, next, progress;
- map nav, guide gates, report, monitor.
A project with no anchors.yaml still runs those that work without one. anchors migrate
is the exception: it has already rewritten files when it reads the configuration, so if it
still does not load it reports so, skips the steps that need it, and asks you to fix it and
run migrate again.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 1858f30 fix(cmd): a broken anchors.yaml is an error, never read as an absent one
0.1.360 2026年10月19日 — 64ビット ZIPARM64 ZIP
リリースノート
dep-honored names a flag whose kind contradicts its import
What reads the map trusts a @dep flag's kind: section-required's depends_on_layer
counts only value dependencies. dep-honored now names two mismatches:
- @dep: on an import of types only. Write @dep[type]:.
- @dep[type]: on an import that runs. Write @dep:.
anchors check --fix already rewrote these flags; until now the gate let them pass.
The ts example's type_import_pattern now also reads braces whose every name is a type,
import { type A, type B as C } from …. If your project declares its own pattern, add the
second branch:
type_import_pattern: '^\s*(?:import|export)\s+type\b|^\s*(?:import|export)\s*\{(?:\s*type\s+[\p{L}_$][\p{L}\p{N}_$]*(?:\s+as\s+[\p{L}_$][\p{L}\p{N}_$]*)?\s*,?)+\s*\}'
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 2813ab9 fix(gate): dep-honored names a flag whose kind contradicts its import
0.1.359 2026年10月19日 — 64ビット ZIPARM64 ZIP
リリースノート
Fixes from a code review: nothing concluded from what was not read
- Project-wide gates check presupposes. A gate of scope: project or batch now checks
its presupposes like any other. With a field undeclared it is pending and names the
field; when the field is waived in dialect.opt_out, it skips.
- An unreadable test or flow leaves testid-consistent pending. When a linked test, a
flow or a neighbour test cannot be read, the gate no longer reports its handles as never
queried. It also no longer passes on the grounds that there is no consumer. It is pending
and names the files. Consumers are read through the run's reader, so check --index reads
them from the git index.
- obligation-honored no longer knows JavaScript identifiers. A token is found at the
edge of an identifier written in letters, digits and _, in any script.
- Generic file names are the project's. dialect.generic_basenames lists the names that
are a role in their folder: a code generated for such a file is made from the folder's
name. Each family's example carries the usual list. A project migrated before this
version has none, so anchors code and anchors new stop treating handler.ts as its
folder until you add:
dialect:
generic_basenames: [handler, index, resource, mod, main, types, route, routes, schema, config]
- The release waits for the CI. The release workflow publishes nothing unless the CI of
the tagged commit passed.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 905c838 fix: project gates presuppose, unread consumers pend, no JS identifier, release waits for CI
0.1.358 2026年10月19日 — 64ビット ZIPARM64 ZIP
リリースノート
Handles built on the caller's head: measured, or pending — never green
A component often builds its testID from the id its caller passes in, for example
`${testID}-option-${opt.value}`. The exposed patterns only read a template that
starts with a literal, so such a unit looked like it exposed nothing. testid-consistent
and section-required then passed without measuring anything.
- The suffix group. dialect.test_handles.exposed now accepts a suffix group: the
segment the unit adds to its caller's head. The ts example reads it, so copy this pattern
into your own:
- 'testID[a-z]*\s*[=:]\s*\{?`\$\{[^}]*\}(?P<suffix>[^`$]*)'
- How a derived handle is checked. The handle becomes *-option-*:
- the component's spec declares it as written, in Test IDs;
- a test or flow queries it by naming the segment after a head, such as :pay-option-a;
- the caller's own handle (:pay) counts as queried through a segment that some unit of
the project derives. A segment that no unit derives counts for nothing, so
:pay-other does not query :pay.
- Handles built from values alone. ${prefix}${key} and ${prefix}-${i} leave the
segment naming nothing. The gates report them as pending, with the file and line, and
never pass them. Give the handle a literal segment.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- ac47f57 feat(gate): a handle built on the caller's head is derived; one of values alone is pending
0.1.357 2026年10月19日 — 64ビット ZIPARM64 ZIP
リリースノート
section-required: a dependency on types only no longer calls for a section
By default, when: {depends_on_layer: [...]} now counts only dependencies on a layer's
values. A component that only imports a layer's types (@dep[type]) receives its data some
other way, so it no longer triggers the requirement. Measured at MIF, this was the case for
22 of 36 charges.
To count type imports too, add kinds:
when: {depends_on_layer: [hook, store], kinds: [value, type]}
The map now records the kind of a depends-on edge: kind: type when the target's types
are its only import. Older binaries ignore the field.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 82e29fa feat(gate): depends_on_layer counts value dependencies; a types-only import asks for kinds
0.1.356 2026年10月19日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: anchors coverage agrees with the scenario-coverage gate
anchors coverage and anchors report now count a spec's requirements the way the
scenario-coverage gate does: only what the spec defines. In particular, they no longer
count as requirements:
- codes the spec only cites, such as handles listed in a Test IDs table;
- rules marked @no-scenario.
Before this fix, the coverage overview could show missing proof that the gate did not
report.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 5ce0f21 fix(coverage): the coverage view counts the requirements scenario-coverage counts
0.1.355 2026年10月18日 — 64ビット ZIPARM64 ZIP
リリースノート
New: section-required, the spec sections your code calls for
A layer can declare the spec sections its code calls for. Each section comes with a trigger
that Anchors already measures:
layers:
component:
requires_sections:
- section: "Test IDs"
when: {exposes: test_handles} # the code exposes a handle
- section: "Data Contract"
when: {depends_on_layer: [hook, store]} # the code depends on these layers
- section: "Navigation"
when: {flag: "@navigates"} # the code carries the flag
- section: "User Messages"
when: {code_matches: 'showToast\('} # your own pattern
The new gate section-required fails a spec that lacks a section its code triggers. The
message names the section, the file and what triggered it.
A section is found by its heading, at any level and in any case. When the unit has none of
what the section covers, write the section anyway and say so and why.
The gate is blocking by default. While you write the sections over an existing tree, set
blocking: false on the gate, or declare a skip with its reason in the commit message.
New: anchors check --show-skips
Lists every indeterminate result (~), grouped by gate and then by reason, the commonest
reason first. Each reason shows its count and the addresses it covers. Use it to tell "does
not apply" from "should have been measured" without reading the gate's source.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 75d39a2 feat(gate): section-required charges the spec sections a unit's code calls for; check --show-skips
0.1.354 2026年10月17日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: mock-stamped finds the stamps anchors stamp wrote
mock-stamped now ties a stamp to a double through the same resolution anchors stamp
uses: the project's aliases, its extensions, then the tail of the path.
It used to compare path suffixes. A double could stay "without stamp" right after
anchors stamp wrote its stamp, in two cases:
- an alias that maps into another folder, such as @mif/shared/x → packages/shared/src/x.ts;
- a directory resolved to its index.ts.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- 4a9d426 fix(gate): mock-stamped finds a stamp by the file the writer resolved, not by suffix
0.1.353 2026年10月17日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: two monitors of one tree keep separate memories
Each anchors monitor now keeps its state in its own file: .anchors/runs/monitor.<ANCHORS_SESSION>.json.
With no ANCHORS_SESSION, the file is still monitor.json.
Two agents watching the same tree used to share one file. Their monitors overwrote each
other's state ("cannot write its state"). An event one monitor reported was also marked as
seen for the other, so that agent never heard it.
A watcher with a session starts with a fresh memory on its first run of this version. It
then says what is going on, once.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- dd0d490 fix(runs): each watcher keeps a monitor state of its own
- c0c3688 test(runs): watchers save at once, each tick after tick, as their monitors do
0.1.352 2026年10月16日 — 64ビット ZIPARM64 ZIP
リリースノート
Fix: mock doubles of another workspace's modules are measured again
v0.1.351 lost doubles whose specifier uses an alias the project declares for a different
workspace. For example, @/lib/x in apps/landing-page, with @/ declared for
apps/mobile/. mock-typed and mock-stamped took them for third-party modules and stopped
measuring them, without saying so.
Now, when the aliased path is not a file of the map, the tail after the alias's prefix is
matched by its suffix.
New: aliases scoped per workspace
dialect.import_resolve.scoped declares aliases that hold only for the importers a glob
matches. Use it for each workspace's own tsconfig paths:
dialect:
import_resolve:
aliases: {"@/": "apps/mobile/"}
scoped:
- files: 'apps/landing-page/**'
aliases: {"@/": "apps/landing-page/src/"}
An importer's scoped aliases win over the global ones for the same prefix.
Everything in the v0.1.351 notes still applies: run anchors migrate once, crossing format 8.
Changelog
- c7c54fa fix(gate): a declared alias that misses the map still names the project's module by its tail
0.1.351 2026年10月16日 — 64ビット ZIPARM64 ZIP
リリースノート
The core reads no syntax — format 8
Breaking. From this release Anchors knows no language by itself. Comments, a language's
lexicon, test names and the shapes some gates used to recognise are now declared in your
anchors.yaml. A gate whose field is not declared is pending and names the field.
anchors doctor shows the family's example to copy.
The map moves to format 8. A map not migrated is refused, with the message to run
anchors migrate.
Upgrade: run anchors migrate once, then commit
anchors migrate crosses format 8 once. Where your project declared nothing, it writes the
examples this binary used to apply by itself, so your check table does not change:
- dialect.comments — the entries covering the extensions of your map's files;
- your family's lexicon — exported_func, param_name, loop, cursor, http_status,
env_read, definition, handle_patterns, log_patterns, guard_patterns,
signature_end, limits, statement_end, set_promise, set_slice. With no family, only
set_promise and set_slice;
- dialect.tests — the family's, or, inside your own, its pattern (when you declared only the
assertion) and the block shape (block_open, block_close, empty_body);
- derived.route_pattern, when you declare route-exists;
- derived.mock_detect, mock_factory and mock_typed, when you declare mock-typed;
- headless: true on a test layer of YAML files (a runner's flows);
- monitor.runners for the tools your suites run with (jest, vitest, go test, pytest, maestro,
playwright, stryker, gremlins).
It prints each field it writes. Run anchors migrate --dry-run first to see the list.
By hand: what migrate cannot write
anchors migrate prints these as ! by hand:
- An extension no example covers. Its files have no header or flag read until you add a
dialect.comments entry for them.
- A dialect.tests.pattern with no title group. The title is what the pattern captures
now, e.g. \bit\(\s*'(?P<title>[^']*)'. Name one group per quote your language uses:
title, title2…
- env-declared with no family. Before, every family's environment reads were read
together. Declare dialect.env_read for your language.
- A section written in flow style — dialect: {family: go}. Migrate refuses to rewrite it.
Turn it into block style and run again.
Not printed by migrate, but worth checking:
- Test names come from derived.files. A test file is paired with its unit by the part
after {{name}} in a template: {{dir}}/{{name}}.test.{{ext}}, {{dir}}/{{name}}_test.go.
This covers the map's identities, duplicate codes, deliver, watch, work and
check --changed. Declare the template your tests follow if you have none.
- Code extensions come from your code layers. Several gates look for a spec's code file:
rule-implemented, unit-complete, the doctrine gates and plan-seeds-valid. They try the
extensions your code layers' patterns end with (*.go, *.{ts,tsx}).
- marker-parity reads only files whose comments you declare.
- No import fallback. These gates no longer read every import/require line by
themselves: proof-crosses-boundary, test-exercises-unit and test-ref-matches-unit.
With no dialect.imports they are pending. anchors doctor offers the family's example.
- Mock aliases. A double's module now resolves through dialect.import_resolve.aliases.
Doubles of your own modules written with an alias (@backend/x) are now charged by
mock-typed and mock-stamped.
- Stamps over a JSX flag. A line holding only {/* @navigates: … */} is a comment as a
whole now. The old reading left {} in the stamped block. A stamp covering such a module
diverges once: renew it with anchors stamp --refresh <module>.
Also in this release
- anchors init writes the family's lexicon and the comments of the extensions it finds.
- anchors doctor offers:
- the comments of the extensions found without an entry;
- the lexicon fields your declared gates presuppose.
- anchors stamp stamps the real module's exports that a double's call names. It reads no
object literal.
- The tests a pattern lists keep their titles as written: the core re
0.1.350 2026年10月16日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 28f00e9 feat(dialect)!: navigation_call is the project's; dotted import paths resolve
0.1.349 2026年10月16日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- d66350d feat(dialect)!: imports and declarations are read by the project's patterns
0.1.348 2026年10月15日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 047417c fix(scan): an identifier the chain reads may hold letters of any script
0.1.347 2026年10月15日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- b7878c6 feat(gate)!: test handles are read by the project's patterns — the core reads no syntax
0.1.346 2026年10月15日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 192be3c fix(gate)!: a fallible call propagates when its pattern's propagates group says so
- 8e2306e fix(ingest): a run drops only the external coverage of its own test layer
0.1.345 2026年10月15日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- e411c2b fix(ingest): a full run drops only the external proof it measured again
0.1.344 2026年10月15日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 1dcdc3b fix(gate)!: a fallible call is named by its pattern's name group — the core reads no syntax
- 969f3b3 fix(gate): a fallible call with a type argument is cited by the name it calls
0.1.343 2026年10月15日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 5909a7b fix(gates,doctor): navigation ends at any heading; a hook with no commit-msg is told
- 1c16524 fix(map): a unit's declared layer tags its spec and feature, so gates scoped by tag find them
0.1.342 2026年10月15日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 578f3a6 feat(guide): anchors guide gates explains the gates — what each measures, if it bars, the way out
0.1.341 2026年10月14日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 53c3b65 fix(map): a proof set aside when a file changes comes back when its content does
0.1.340 2026年10月14日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 2ae16ed fix(code): a suggested code is reserved, across the repository's worktrees
0.1.339 2026年10月14日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- e6bafe3 fix(monitor): the heartbeat says the CPUs' use, and HIGH needs them busy, not the load alone
0.1.338 2026年10月13日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 1fc2b87 fix(monitor): a holder's edits in its worktree are activity, and its quiet tasks go in one line
0.1.337 2026年10月13日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 9ef79b6 fix(monitor): a run still going in another worktree keeps its holder active
0.1.336 2026年10月13日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 4698cd8 fix(code): a file's own code is taken — --check, code list and suggestions see it
0.1.335 2026年10月13日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- e6a8970 feat(progress): move takes a task to the stage it arose in, and drop removes an empty stage
0.1.334 2026年10月13日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 40551ef feat(progress): rename gives a demand, a stage or a task a new title, keeping its history
- c39d65f feat(progress): unplanned work is recorded where it arises, marked, and the flow says so
0.1.333 2026年10月13日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 31d6028 feat(test): a test that ran out of time under load runs once more, and says so
0.1.332 2026年10月13日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 7d774f9 fix(monitor): a held task's silence counts activity in every worktree, since the monitor watched
0.1.331 2026年10月13日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- bac2051 feat(monitor): a held task with no activity is said, and subagents are runs
- c762144 feat(monitor): the summary goes out when it changes; --level action says only what needs a reaction
- 6023d01 fix(test): an impact path over the command-line ceiling runs in batches
0.1.330 2026年10月13日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 45c6dd4 feat(monitor): the agent copies the progress summary verbatim, total included
0.1.329 2026年10月12日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- d859d3d feat(monitor): a line apart asks the agent to show the user the progress summary
- 38e550b fix(check,monitor): Anchors' own files have nothing to confront, and that is no failure
0.1.328 2026年10月12日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 1273f88 feat(monitor): says the progress summary with what it says — --summary or monitor.summary
- f410bf8 feat(progress): one anchors.progress.md of weighted demands, written only by anchors progress
- 906633b fix(lock): the release insists when Windows refuses to delete a lock being read
0.1.327 2026年10月12日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 2e5e9fd fix(fix): a file of a declarative layer gets its code and its layer
0.1.326 2026年10月11日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- a642bef feat(guidance): the first canonical layer — how the project says it is worked on
0.1.325 2026年10月11日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- a01fb06 feat(init): AGENTS.md and CLAUDE.md point to CONTRIBUTING.md, which asks to watch long work
0.1.324 2026年10月11日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 0e74046 feat(hooks,review): hooks speak the project's language; each review item is a finding
0.1.323 2026年10月11日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 6a93121 docs(site): the affected-tests command in the confronting family
- 90167ed feat(chain,test): OCaml modules become edges, and the tests a change can break are listed
0.1.322 2026年10月11日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 2e66950 fix(init,config): a broken anchors.yaml is fixed, ignored folders are no layers, OCaml is read
0.1.321 2026年10月11日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 14ef571 feat(freeze): a frozen part is kept as it is, quiet, and told when touched (FRUNF)
0.1.320 2026年10月10日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 489db79 perf(gate): a check over MIF from 14m29s to 5m01s, with the same verdicts
0.1.319 2026年10月10日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 21daa7c fix(map): a capture test depends on the baselines it compares against, not its neighbours'
0.1.318 2026年10月10日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 3986f07 feat(scan): a line that is only a comment is no evidence
0.1.317 2026年10月9日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 54b26c1 feat(map): a test's evidence follows its asserts, not the wiring that gets it there (EFTAV)
- f8355b1 feat(map): what has no side effect on the tests stales no proof — @navigates and @dep[type]
- b450907 fix(monitor): a report with no test is no result
0.1.316 2026年10月9日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 065df0a fix(monitor): only background commands are the hook's runs, read by their first line
0.1.315 2026年10月9日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 880998c feat(monitor): the heartbeat says the load against the CPUs (load=8.4/10)
0.1.314 2026年10月9日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 6ba19fe feat(install-hooks): --agent --user puts the agent hook in the user's own settings
- 2dd918c fix(monitor): a blocked check records its exit, and the shell around a run is no run
0.1.313 2026年10月9日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- af7a15c feat(monitor): anchors monitor — one line when a long process starts, stalls, dies or ends (PRMNP)
0.1.312 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 3224730 fix(map): refreshing revs after a run carries the evidence a flag-only edit left proven
- cd9cf22 fix(stamp): --refresh takes every file given after it, not only the first
0.1.311 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- f8d353d chore(map): the committed map carries the evidence revisions (EBWIP)
- cf1b687 feat(map): a proof goes stale when what it proves changes, not when the file does (EBWIP)
0.1.310 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 22cf6b2 fix(scan): a bare @dormant keeps the navigation flag, as a plain edge
0.1.309 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 734ec3c feat(gate): a dormant navigation annotates its call and is no edge (NCGNV-B08)
0.1.308 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 45c836a feat(gate): flags, code and features confront their repeats (DPDCD-W03)
0.1.307 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- ccd71a3 feat(gate): a rule of another unit a spec cites lives in the product (DOOSD-D05)
0.1.306 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- ba34e9d feat(map): kinds of dependency — what the code reaches that is no import (DOOSD-W04)
0.1.305 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- a12bad8 feat(spec): the spec declares no dependency; migrate takes the tables out (DOOSD-W02, W03)
0.1.304 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 4cf5b62 fix(gate): a test lists in @used-by only when the project puts its tests in the chain
0.1.303 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 1601efd feat(gate): a component's navigation counts for the screens that render it (NCGNV-B07)
- dbd289b feat(gate): the dependency chain reaches every artifact that imports (DOOSD-W01)
- d711eae fix(touch): a touch run on its own carries the evidence in the map (HDTHD-B16)
0.1.302 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- ef6132a fix(gate): a stamp reads out a flag in a block comment inside a line (MCSTM-B22)
0.1.301 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 7e84274 fix(gate): the navigation fixer never writes into JSX text; suites run without the hook's repository
0.1.300 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- cf59635 feat(gate): the spec catalogue confronts its repeats (DPDCD-W02)
- c5dddd1 fix(fix): a repair carries only the evidence that held before it (EDSTD-B18)
0.1.299 2026年10月8日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 53ab9a4 feat(gate): each gate confronts the repeats of what it declares; rule codes first (DPDCD-W01)
- 0ac763b fix(gate): a mock stamp reads its module without the chain's flags (MCSTM-B21)
0.1.298 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- a1ec29c fix(gate): contract-impact answers a rule where it lives (CTRIM-B06)
0.1.297 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 24125fc fix(new): a rule letter keeps one home in the spec, and no code is defined twice (NWARN-B21)
0.1.296 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 688943a feat(map): a test that asserts a navigation goes stale when its Out row changes (DNDDP-W05)
0.1.295 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- d243a63 fix(docs): the navigation page places an edge that lands on the screen's spec
- 604683d test(map): flagged imports feed the capture's closure and the change's impact (VRCPT-B09)
0.1.294 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- a75c9de feat(docs): the navigation map and the dependency chain as pages; their guides (DNDDP-W04)
- 1f63b62 fix(gate): a re-export declares what it lists; an inline import brings its member (DCGDP-B06)
- 75ee28a fix(scan): a destructured dynamic import brings the names it destructures
0.1.293 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 7606fa3 feat(gate): the navigation chain — annotated, matches-spec, symmetric, reachable (DNDDP-W03)
0.1.292 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 5287b34 fix(fix): one git snapshot per run of the fixers; the site pages of the chain's gates
0.1.291 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- d0b6a23 feat(gate): the dependency chain — dep-declared, dep-honored, used-by-declared (DNDDP-W02)
0.1.290 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- f96b80a feat(map): the code's flags — @dep, @used-by, @navigates — read and mapped (DNDDP-W01)
0.1.289 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- b131fea feat(new): the screen preset writes the load states and the load's failure (DFLSR-W04)
0.1.287 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 5dbca21 feat(gate): each fallible call has its handling beside it (DFLSR-W02)
0.1.286 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- fe460aa feat(gate): a retired rule stays as a tombstone; a change's revision answers its impact
0.1.285 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- edf179e feat(gate): a unit that can fail names how — fallible sources (DFLSR-W01)
0.1.284 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- d6c7f81 feat(gate): data states as requirements, opt-in; their passing tests are kept
0.1.283 2026年10月7日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- b073c32 fix(new): a rule section takes its letter's own title, not the shared one
0.1.282 2026年10月6日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 4532dab fix(gate): a code an open question cites as context does not close it
0.1.281 2026年10月6日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 27c6099 fix(gate): an open decision is no requirement
0.1.280 2026年10月5日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 1145440 fix(evidence): keep-evidence brings back the suites a later run did not replace
0.1.279 2026年10月5日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- d450817 fix(gate): a @no-vr exempts its own row, not a code its reason mentions
0.1.278 2026年10月4日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- f430d06 fix(gate): a mechanical change to a plan asks for no revision
0.1.277 2026年10月4日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- f75f7a2 fix(migrate): names a code heads keep following it, and binaries stay untouched
0.1.276 2026年10月4日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 0baf8ac fix(migrate): rename a code only where it is cited as one, and fast
0.1.275 2026年10月4日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 4ce00ab fix(map): a file's unit is the one it refs, not its own code
0.1.274 2026年10月4日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 7ec9af1 fix(migrate): format 7 leaves nothing stale, and old codes stay aliases
- 05a3914 fix(migrate): widened rule codes also follow in files the project does not govern
0.1.273 2026年10月4日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 531d1ea feat(migrate): format 7 — every governed file gets a code of its own
0.1.272 2026年10月4日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- d51c8f7 feat(map): VR evidence follows what changes the look, and a file's own code
0.1.271 2026年10月4日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 5e260dc fix(gate): a spec section is found at any heading level
0.1.270 2026年10月4日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- a33bacb feat(env): environment variables declared in the spec, confronted with the code, listed in docs
0.1.269 2026年10月3日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 27b4558 feat(gate): validations lead to states, errors show messages, messages are captured
0.1.268 2026年10月3日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 4e506c2 feat(api): API spec sections, and the project's OpenAPI compiled from them
- 2518239 feat(gate): contract-tested — an API unit is proven against the compiled OpenAPI
- 45e085e feat(gate): the coherence of an API spec, and its error codes in the code
- 7e20b61 feat(testsig): a green capture of a state proves the state
0.1.267 2026年10月3日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 3c5db35 fix(mutation): a file that gives no mutant is measured, not pending forever
0.1.266 2026年10月3日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 2e8ae7a feat(gate): exempt a state with no visual value with @no-vr: <reason>
0.1.265 2026年10月3日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- d0b2351 feat(map): a VR test captures its unit, so a screen's change stales its capture
0.1.264 2026年10月3日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 6edb40a fix(gate): a VR code carries its state, so scenario-identity reads each state apart
0.1.263 2026年10月3日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 500a2bf ci(gates): gofmt is a blocking gate, so the pre-commit catches what the CI did
- ef4513f feat(gate): the four questions of visual regression, one VR scenario per state
- 3719736 feat(gate): vr-states-covered — every state of a screen or component has visual regression
0.1.262 2026年10月3日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- d3af065 fix(check): a gate the run leaves out is still declared in the catalog line
- 5c1e869 fix(check): suggest only the catalog gates that would measure here
0.1.261 2026年10月3日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- ce40b7c feat(report-bug): agents report bugs in Anchors itself to co2-lab/anchors
- bf9c985 feat: issue forms for bugs and features, and --upstream writes the bug form
0.1.260 2026年10月3日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 865e74f fix(mutation): a file where no mutant ran has no score, and the gate does not apply
- c207dd5 fix(sitecli): a re-run leaves the command pages as they are
0.1.259 2026年10月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- b3eb52e ci(site): deploy to the production branch, main
- da25aa5 docs(site): one page per command and per gate, generated from the code
- 1f6ca7a feat(escalate): --upstream reports a bug in Anchors itself to co2-lab/anchors
- 2a4d937 fix(daemon): on Windows, an exited process is not alive
- 70bd10d fix(map): a later release restamps the map, so the stale-binary warning clears
0.1.258 2026年10月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 5641b10 docs(site): the unit, review, the header everywhere, init without presets
- 4754860 feat(gate)!: every governed file carries the header, and --fix writes it from the map
0.1.257 2026年10月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 7e6a698 feat(init): seed what relates to the layers; doctor and check name the rest of the catalog
- fc2eb38 fix(gate): a -- header, a YAML comment that is no heading, a rule charged once
0.1.256 2026年10月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 6eef3e2 feat(gate): review — a gate's targets marked for a second look, apart from its verdict
- a801ebe fix: the ref-resolves message names header-valid; test repos run no detached gc
0.1.255 2026年10月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 2e0543f fix(gate): manual gates stay manual, and a gate says the field it presupposes
0.1.254 2026年10月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 1ec76dc refactor(gates)!: the gate steps are Go, not shell — and init says to do the same
0.1.253 2026年10月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 7397849 fix(gate): a test's fake is not a copy, coverage floors per gate, Go JUnit finds its file
0.1.252 2026年10月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 59f7af4 fix(docs): rules in every form, sections in any language, pages in the project's
0.1.251 2026年10月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- a48e14d fix(dialect): the go family sees an error in a field and a sentinel error
0.1.250 2026年10月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 10ecaf2 feat(gate)!: the triad is the unit — triad-complete becomes unit-complete
0.1.249 2026年10月2日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 18c139c feat(init)!: read the project's folders and language, propose no structure, seed CONTRIBUTING.md
0.1.248 2026年10月1日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 3e49e7c feat(config): a project-wide severity, the default of its blocking gates
0.1.247 2026年10月1日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 8ccac11 fix(gate)!: with no severity, a blocking gate blocks only its failures
- 94d4e41 fix(gate): an open decision informs by default, as any divergence
0.1.246 2026年10月1日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- afd955f fix(coverage): a suite that ran none of a file leaves its coverage
0.1.245 2026年10月1日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 2f3ccab feat(gate)!: each verdict level takes a state — block, inform or ignore
0.1.244 2026年9月30日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- b50c93c fix(coverage)!: a rule is proven only when each of its variants is
0.1.243 2026年9月30日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 8a6ce4d fix(cli): every command that takes files reads a list the same way
0.1.242 2026年9月30日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 8c39f35 docs(guide): a spec is written in four passes, and reviewed
- 72d3c5a feat(verify): the pre-commit compiles and stages the docs
0.1.241 2026年9月30日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- b7be4e7 fix: an it.each table closes into its test, and every named file is answered
0.1.240 2026年9月30日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- bbd14c7 fix(test): the selection reads the map as a build would leave it now
0.1.239 2026年9月30日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 065dcc1 fix(touch): named files narrow the touch to them
0.1.238 2026年9月30日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 71f3ca7 fix(gate)!: remove workdir: index and the copy of the index
0.1.237 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- add9b1f fix(gate): under --index the tests are listed from the commit
0.1.236 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- da64771 feat(gate): a run: gate can run in a copy of the index
0.1.235 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 55b1427 fix(scan): the index reader confronts the tree at each read
0.1.234 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 157ea1e fix(gate): the index copy is passed with forward slashes
- 142490a fix(gate): under --index a command gate reads the commit's content too
0.1.233 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 3a5ca35 fix(check): a commit is judged with its own map, and a file not in it is not judged
0.1.232 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 0e1715e ci: macOS and Windows block, now that they pass
- 6606708 ci: test on macOS and Windows, informational until clean
- 48d1a19 fix(new): the spec an identity is read from is named with /
- 6af43d2 fix: Windows paths, a crashed map writer, and the budget's process tree
- fcb721d fix: on Windows, a lock being released is busy, and init does not wait on a console
- f9429fc fix: paths keep / on Windows where the map and the reader expect it
- 632d4ca fix: run on Windows and with only git and gh installed
0.1.231 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 9f0e784 feat(stamp): handed a test, the refresh names the modules to refresh
- 59dc617 fix(hook): the commit's map keeps what HEAD proved, and the tree keeps its own
0.1.230 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 7ce19fc fix(evidence): a kept declaration is not a measurement, and the hook dates by the index
0.1.229 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 34f00f5 fix(check): a commit is judged by what it records, not by the tree
- cdb65bf style: gofmt the index reader and the verify args
0.1.228 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 2ad73cb feat(quality): keep-evidence — a change that proves nothing new keeps what was proven
- 159db15 fix(scan)!: the header is the block at the top of the file
0.1.227 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 0680c99 feat(gate): header-layer-declared — a header's layer is one the Estrutura has
0.1.226 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- bd66d2d fix(docs): only a layer that runs code is out of every container
0.1.225 2026年9月29日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 993276d fix(docs): a container's layer with no spec is said, not an error
0.1.224 2026年9月28日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 03f2475 feat(gate): branches no test takes, and Examples rows no test runs
0.1.223 2026年9月28日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 4f04e7b feat(gate): a test asserts, and reaches the unit it says it tests
0.1.221 2026年9月28日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 95e62db feat(gate): contract-impact — a changed field names the rules that use it, and their tests run
0.1.220 2026年9月28日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 04136c4 feat(gate): rule uses reach the code, and presentation validations are confronted
0.1.219 2026年9月28日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 1afd89f feat(gate): every test citing a code, copied scenario bodies, one test file per unit
- db79b37 fix(mutation): a run measures only what the gate confronts
0.1.218 2026年9月28日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- b91517e feat(map): a new file enters the map at once, reading only its unit
- 95cb2d3 fix(map): the edges are in a total order
0.1.217 2026年9月28日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- e3ced99 feat(spec): tie each rule to what it uses — Validations, Presentation validations, Rule uses
0.1.216 2026年9月28日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 43117e2 feat(vocabulary)!: the letters of plans, flows and actions in English, by migration
- d4f0336 fix(new): the plan template follows the project's language, with its own title
- 0086909 fix(plan): a phase's dependency is read in any supported language
0.1.215 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- e38834c chore(map): rebuild after the hook dated the staged specs
- cb0a68b fix(map): each command applies only its change, to the map re-read under a lock
0.1.214 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- cb0e5e9 fix(changelog): a Bug block followed by a Co-Authored-By block is still the footer
0.1.213 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- e0ff969 ci(site): build the site on Node 22 — Astro refuses Node 20
- e7a661e fix(ingest): a run's proofs are stamped with the tree's revs, not the map's
- 3c48022 fix(mutation): a mutation result keeps its own rev; coverage no longer makes it current
- 203c2d1 fix(suite): a report stamped by a coarse clock just before the run is this run's
0.1.212 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- efb9c77 feat(changelog): anchors changelog builds the technical changelog from the commits
0.1.211 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 45bc75a feat: a fix is its own commit, and a bug — a defect that shipped — carries a Bug: footer
- 98fe1e9 fix(dialect): the TS family recognises catch { without a binding
0.1.210 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 85a1ab5 chore(mutation): measure the budget and selection files; the coefficient can be raised
- 22b8526 feat(gate): a test of a declarative unit is not charged a feature
0.1.209 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 690c6a2 chore(mutation): ingest the re-measurement of the six changed packages
- 624ffd6 fix(budget): a cut batch gets a TERM and a grace before the KILL
0.1.208 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 6686717 fix(suite): a suite declares its files with paths:, and is handed no other
0.1.207 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- dea2481 feat: runs take only what is stale and below the minimum, with flags to take the rest
0.1.206 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- e1cb635 feat(gate): distrust mutation scores measured under load; declare targets with nothing to measure
- 4eb9c18 feat: record each file's run time, and run fastest first within a --budget
- 287732a fix(mutation): a mutant that times out on a slow rebuild is not a killed mutant
0.1.205 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- b115b9c chore(gates): drop plan-doctrine-exists, and put the docs-covered comment back on its gate
- 3d06583 chore(gates): header-valid blocks, and its comment says what it measures now
- 736bc4f chore(gates): promote the six clean informative gates to blocking
- 4dfbb03 chore(mutation): measure every package, not only internal/gate
- cd1ab42 docs(spec): test-level-codes closes its open decisions with none
- 8bace69 feat(gate): no-duplication reads jscpd's report and judges each file
- 570ba3a feat(init): header-valid joins the canonical catalog, on specs and features
- 4e7f36d feat: the project says how its tests are written, and which test files are support
- 34a7a85 fix(gate): a file the tests source lists no test in is read as without a source
- 14c34c9 refactor: calibrate jscpd for this repository and remove its two real duplications
0.1.204 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- a3845c8 feat(gate): test-level-codes — each test level accepts only the codes it declares
0.1.203 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- a6c9975 fix(identity): a rule's VR baseline is read by its unit code
0.1.202 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- c37ef9d chore(map): the mutation signal after the surviving mutants were killed
- e81a9ec fix(failure): the failure gates read Go as written; specs say why each failure is absorbed
- 006a857 fix: the ~95 bugs the triad specs surfaced
- b972679 test(gate): kill the surviving mutants of four gates, and fix route-declared's section reader
0.1.201 2026年9月27日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 1dfdf7a feat(spec): every code file of Anchors under a triad
0.1.200 2026年9月26日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 935fe43 build(test): the unit suite also reports line coverage
- 1dadef4 fix(gate): close the last failure-declared findings, and the minor bugs of the review
0.1.199 2026年9月26日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- c59e596 feat(spec): rule aliases, and real failure rules for the gates
- f267b45 fix(gate): nine gates that answered for what they never read
0.1.198 2026年9月26日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- d02a15c feat(gate): close the two open decisions; docs-fresh no longer trips on the header date
0.1.197 2026年9月26日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- 543d029 fix(check): clear the check --all failures, and promote the gates to blocking
- 129c550 fix(gate): resolve feature-test-match divergences and catalog failure rules
- 4257651 fix(ingest): a full run drops the proof that reports from outside the repo left
- 70e7dd3 fix(review): address third review points across specs, tests, and map
0.1.196 2026年9月26日 — 64ビット ZIPARM64 ZIP
リリースノート
Changelog
- e9cf61b fix(ingest): a partial test run keeps the proof of what it did not run
- 58cdc1e fix(touch): only the header at the top of the file is dated
0.1.2 2026年8月28日 — 64ビット ZIPARM64 ZIP

プログラミング言語・SDKで最も人気

すべて表示
  1. Go Programming Languageプログラミング言語・SDK
  2. LLVMプログラミング言語・SDK · LLVM
  3. Denoプログラミング言語・SDK · Deno Land Inc.
  4. CMakeプログラミング言語・SDK · Kitware
  5. Yarnプログラミング言語・SDK · Yarn Contributors
  6. pnpmプログラミング言語・SDK · pnpm
  7. Ccacheプログラミング言語・SDK · Joel Rosdahl
  8. sbtプログラミング言語・SDK · Lightbend, Inc.
  9. BOINCプログラミング言語・SDK · Space Sciences Laboratory, U.C. Berkeley
  10. Dart SDKプログラミング言語・SDK · Google

co2labのほかのソフト

すべて表示
  • PolvoIDE・コードエディター · co2lab

タグ

共有