crewhaus च्या जुन्या आवृत्त्या डाउनलोड करा

**Answers that carry their proof.** A spec can now require evidence for what
its harness says. The new `provability:` block sets a floor on a four-rung
ladder. At `cited` every claim names a source. At `grounded` the run read
that source, and it says what the claim says. At `checked` a tool the run
called computed the value, and a pure tool is run again to confirm it. At
`proven` a proof checker accepted a proof of the claim. Before an answer is
served, a deterministic gate checks each claim against what the run
recorded. An answer that falls short is retried with a note naming what
failed, handed to a stronger model, served with a visible ✗, or refused, as
the spec's `on_fail` says, and the verdict travels with the answer as a proof
bundle. The gate is off until a spec turns it on：a spec without the block
runs as before, apart from the changes under **Before you upgrade** and the
fixes under **Fixed**. Specs stay `version：2`, with nothing to migrate. The
design is in
[design/0.8.0-provability-release.md](design/0.8.0-provability-release.md).

**0.8.0 also ships the 0.7.1 hardening, which makes the 0.7.0 tools safe to
grant.** 0.7.1 was never tagged, so its notes are part of this entry：under
each heading below the provability release's entries come first, and the
hardening's follow as they were written. An audit of 0.7.0 found that
its new tools did less than the release notes said, and could do more than a
spec allowed. A permission rule could be dodged by spelling a call another
way. A tool could write through a symbolic link planted in the workspace, or
hang on a crafted file or pattern. The model could choose which environment
variable a tool sent as a credential. And `tool_config` never reached many of
the tools that read it, while shapes other than `cli` could not use the new
tools at all. This release fixes what the audit confirmed, apart from the
items under **Known limitations**. Most specs need no change. To upgrade：use
Bun 1.3.14 or later; run `crewhaus lint` and `crewhaus compile --strict` on
each spec, which point out most rules, tool lists and `tool_config` blocks
this release reads differently; read **Before you upgrade** under Changed;
then recompile your bundles, since a bundle uses the packages of the CLI that
built it.

### Added

#### Provability

• **The `provability:` block**, on cli, channel, managed and research specs.
`provability：grounded` sets the floor and takes every default. The object
form is strict, so a misspelt key fails the build, and the error names each
key that is wrong. Beside `level` it takes `on_fail` (`retry` by default,
`halt`, `note`, or `escalate`, which needs `agent.model_pool`),
`max_retries` (default 1), `threshold` (the share of claims that must meet
the floor, default 1.0), `scope` (`session`, or `turn` for evidence from the
current turn only), `assumptions` (`allow` or `forbid`) and
`max_assumptions`, `render` (`footer`, `none` or `json`), and four
sub-blocks：`sources` (host-glob `allow` and `deny` lists, `bibliographic`
to accept free-text references, and `resolve：head` for a reachability
check), `grounded` (`coverage`, default 0.8; `window_tokens`, default 60;
`independent`, the distinct origins each claim needs, default 1),
`execution` (the tools whose results count as evidence, by default
`[all-compute, all-verify, -goldenUpdate]`, the read-only computation and
verification tools; `replay`; `max_replays`) and `formal` (`checkers`, by
default `[builtin, smt]`, with `builtin` added when a list leaves it out; an
`image` for the lean4 checker; the Lean pre-scan's `forbid` list;
`timeout_ms`). An optional `judge.model` may excuse a paraphrase the
grounded check missed, and nothing else; it is metered as a judge under
`budget.judge_share`, and `judge.allow_self_judge：true` accepts a judge
that is also a serving model. On research, `on_fail` defaults to `note`,
and `retry` or `escalate` lower to `note` with a warning. A crew spec
accepts the block with an `accepted-but-unwired` w

**Tools a harness can run without spending a token.** A CrewHaus harness could
call 22 builtin tools; it can now call 549. They read chains and ledgers, drive
git and package managers, gate a page before it publishes, operate a fleet, and
answer the hundreds of small questions a run used to spend a model call on —
almost none of them makes one, and a spec turns them on a capability group at a
time rather than a name at a time. Two of the fixes below are worth reading even if
you upgrade nothing：an SSRF classifier that compared address text, found by
auditing all ten copies of it, and a daemon status page that had been binding
the wildcard rather than loopback.

### Added

• **Fifty-seven new `@crewhaus/tool-*` packages** (#453, #454, #458, #459,
#462, #463, #465, #466, #467, #469, #470, #471, #472). Each is an ordinary
builtin — it takes a typed input, does the work in-process, and returns a
result — so the work it does costs no tokens and cannot drift between runs.
Two of the 549 are the exception and say so in their own descriptions:
`ImageGenerate` calls a remote image model and `WebSearch` calls a paid
search API. Every other one is local work. Broadly:

• **Pure computation, no I/O at all.** Text, structured data (JSON, YAML,
TOML, CSV, XML), encoding and hashing, dates and schedules, maths and
exact money, schema validation, control flow, and tabular intake:
`tool-text`, `tool-data`, `tool-encode`, `tool-datetime`, `tool-math`,
`tool-schema`, `tool-flow`, `tool-table`.
• **The working tree and the toolchain.** Files and archives, processes,
git, GitHub and GitLab, SQLite, documents, HTML, packaging and lockfiles,
code intelligence, and the gates you run before a change lands —
`tool-fsx`, `tool-proc`, `tool-git`, `tool-codehost`, `tool-sql`,
`tool-docs`, `tool-html`, `tool-pkg`, `tool-code`, `tool-verify`,
`tool-changeset`, `tool-buildperf`, `tool-supplychain`, `tool-registry`,
`tool-containers`, `tool-distribution`.
• **What a harness remembers and what it reports.** Durable state, secrets
handling, messaging, observability and cost：`tool-state`, `tool-secure`,
`tool-secrets`, `tool-notify`, `tool-obs`, `tool-media`.
• **Money and chains, reading only.** ABI encoding and EIP-712 digests,
live EVM reads, token and DeFi queries, a hash-chained local ledger,
e-invoices, payment files and counterparty checks：`tool-onchain`,
`tool-chainread`, `tool-chaincall`, `tool-token`, `tool-defi`,
`tool-ledger`, `tool-einvoice`, `tool-kyc`, `tool-money`,
`tool-objectstore`.
• **The operator's own machine.** System and network facts, scheduled jobs,
filesystem watches, the system package manager and the desktop:
`tool-host`, `tool-hostfs`, `tool-cron`, `tool-pkgmgr`, `tool-desktop`.
• **CrewHaus operating itself.** Specs, evals, datasets, approvals, harness
lifecycle, fleets, deployments, routing and discovery：`tool-specops`,
`tool-evalops`, `tool-dataset`, `tool-approvals`, `tool-lifecycle`,
`tool-fleet`, `tool-deploy`, `tool-routing`, `tool-discovery`,
`tool-crewhaus`.

**What these tools deliberately will not do matters as much as what they
do,** because each of these was a capability that would have been easy to
add and wrong to ship:

• **None of these tools signs or sends a transaction.** They read. No
schema among them takes a private key, and each of the five chain
packages refuses `eth_sendTransaction` and `eth_sendRawTransaction` at
the RPC seam by assertion rather than by omission. (The separate
`@crewhaus/tool-evm-tx`, which does sign, is wired per shape and is not
one of the 549.)
• **Nothing moves money.** `PaymentFileBuild` writes a payment file and
`InvoiceRender` renders an invoice; transmitting either is a human's job
through their own bank. No schema takes a banking credential.
• **Nothing acquires privilege.** `PackageInstall` never runs `sudo`,
`doas`, `runas` or `pkexec`, and no schema has a password field. A
manager that needs root refuses and hands back the exact command an
operator would run; Homebrew, which doe

## What's Changed
• docs：chvm is the recommended way to install crewhaus by @StudioMaxIO in https://github.com/crewhaus/factory/pull/419
• chvm is at crewhaus/chvm, and covers Windows by @crewhaus in https://github.com/crewhaus/factory/pull/420
• chvm installs in one command now by @crewhaus in https://github.com/crewhaus/factory/pull/421
• fix(crew-orchestrator)：forward per-role model_pool/model_tiers/model_fallbacks to the role loop by @crewhaus in https://github.com/crewhaus/factory/pull/422
• feat(runtime-core)：budget substrate — gate every model call, pool-aware degrade, toolsInPlay union, streaming model latency by @crewhaus in https://github.com/crewhaus/factory/pull/423
• feat(trace)：role/stage/profile attribution on model events, model_stage + model_directive kinds, durable routing event-log kinds by @crewhaus in https://github.com/crewhaus/factory/pull/424
• feat(model-plan)：new @crewhaus/model-plan package, strongest sentinel, requiresModelFeatures, adapter effectiveParams by @crewhaus in https://github.com/crewhaus/factory/pull/425
• feat(eval-judge)：judge and compaction spend on the run bus, budget.judge_share, run-spanning workflow meter by @crewhaus in https://github.com/crewhaus/factory/pull/426
• feat(observability)：OTel model attributes and metrics, escalation/judge/floor alerts and SLO, session pin/models fields, sessions tail and export by @crewhaus in https://github.com/crewhaus/factory/pull/427
• feat(spec)：models: profile registry, $profile references, hybrid pool blocks, escalate, judge panels, per-node/sub-agent routing by @crewhaus in https://github.com/crewhaus/factory/pull/428
• feat(ir,compiler)：IrModelProfile, $profile resolution at every slot, modelPlanIntegrity pass, capability/sunset validation by @crewhaus in https://github.com/crewhaus/factory/pull/429
• feat(crew)：per-candidate profile enrichment on roles and crew.routing.model for the llm router by @crewhaus in https://github.com/crewhaus/factory/pull/431
• refactor(model-service)：wireModels composition root wrapping today's emitter routing fragments byte-identically by @crewhaus in https://github.com/crewhaus/factory/pull/430
• feat(tool-consult)：Consult + Escalate tools, TrustOrigin consult, narrowRuleSet, interpreter parity by @crewhaus in https://github.com/crewhaus/factory/pull/432
• feat(migration)：MIGRATION_1_TO_2 via SpecEdits, irreversible guard, comment-preserving upgrade and migrate-all, --hoist-models, per-release upgrade notes, pricing-feed sunsets by @crewhaus in https://github.com/crewhaus/factory/pull/433
• feat(runtime-core)：per-candidate plan table — request params, subset tool advertisement, dispatch gate, per-candidate permissions/rate-limits/tool_config, volatile overlay by @crewhaus in https://github.com/crewhaus/factory/pull/434
• feat(runtime-core)：preRoute decision phase — directives with session pin, rules, capability eligibility, classifier hook, budget hints, pre-model mutate by @crewhaus in https://github.com/crewhaus/factory/pull/435
• feat(model-service)：guide, shadow, and committee side-call closures with nested single-turn isolation by @crewhaus in https://github.com/crewhaus/factory/pull/437
• feat(sub-agents)：per-sub-agent routing, projectParentHandle, Task profile allowlist, inherit_routing, child spend re-published by @crewhaus in https://github.com/crewhaus/factory/pull/438
• feat(runtime-core)：cascade and draft-verify — runOneTurn force, evaluation.on_fail escalate, clean-prompt snapshot, deferred quality fold by @crewhaus in https://github.com/crewhaus/factory/pull/436
• feat(spec-patch)：wildcard segment, model_pool structural rule, the 0.6.0 OPTIMIZABLE_PATHS delta with fixtures, Hangar spec-edit tiers, lint/doctor checks by @crewhaus in https://github.com/crewhaus/factory/pull/439
• feat(model-router,routing-store)：RouteSignals + hint input, policy union, scoped arms with backoff, per-candidate chains, v2 lines, priors, floor, route freeze by @crewhaus in https://github.com/crewhaus/factory/pull/440

## What's Changed
• docs(packages)：refresh release facts and drop the stale delete-bun.lock advice by @crewhaus in https://github.com/crewhaus/factory/pull/416
• feat(hangar)：M5 Advisor — unified alerts/suggestions with quick actions, decision ledger, reports, issue inbox, and Library curation by @StudioMaxIO in https://github.com/crewhaus/factory/pull/417
• chore(release)：crewhaus 0.5.8 by @StudioMaxIO in https://github.com/crewhaus/factory/pull/418

## New Contributors
• @StudioMaxIO made their first contribution in https://github.com/crewhaus/factory/pull/417

**Full Changelog**：https://github.com/crewhaus/factory/compare/v0.5.7...v0.5.8

## What's Changed
• docs(changelog)：correct the blast radius of the 0.5.5 zod gap by @crewhaus in https://github.com/crewhaus/factory/pull/412
• fix：default llm_judge 400s on Claude 5-family models — omit the pinned temperature where the API rejects it by @crewhaus in https://github.com/crewhaus/factory/pull/414
• chore(release)：crewhaus 0.5.7 by @crewhaus in https://github.com/crewhaus/factory/pull/415

**Full Changelog**：https://github.com/crewhaus/factory/compare/v0.5.6...v0.5.7