Specrew v0.38.0

Feature 184 — Full Antigravity Refocus. BETA (no stable or full-parity claim). Builds on the 0.37.0
(Feature 174 — Hook-Driven Session Bootstrap) base.

Added

• Persistent host coordinator instructions at specrew init (Feature 184): specrew init merges a
  Specrew-owned, delimited coordinator section — carrying the anti-raw-specify.exe workflow guard — into
  each supported host's manifest-declared InstructionsFile (AGENTS.md, CLAUDE.md,
  .github/copilot-instructions.md), preserving user content byte-for-byte; specrew update refreshes it and
  specrew start heals a missing or stale section. Content comes from one packaged fragment in the module
  FileList, and the shared delivery core is host-neutral (manifest-driven, host-coupling-firewall-guarded).
  This closes the last Antigravity parity gap: a cold-launched agent comes up as the governed Specrew
  coordinator instead of a blank session. Real-host-validated for strong models (Opus 4.6, Claude).

Notes

• Weak-model caveat (FR-017): Gemini Flash facilitated the governed workshop competently but then
  self-authorized lifecycle boundaries — the coordinator must be a strong model until a deterministic gate
  (Proposal 180) lands. No full or verified Antigravity-parity claim; release carry-forwards
  (beta-before-stable, MigrateLegacyTopLevelEventMap legacy-upgrade validation, machine-local agy
  evidence) remain open.

Specrew v0.38.0-beta1

Feature 184 — Full Antigravity Refocus. BETA (no stable or full-parity claim). Builds on the 0.37.0
(Feature 174 — Hook-Driven Session Bootstrap) base.

Added

• Persistent host coordinator instructions at specrew init (Feature 184): specrew init merges a
  Specrew-owned, delimited coordinator section — carrying the anti-raw-specify.exe workflow guard — into
  each supported host's manifest-declared InstructionsFile (AGENTS.md, CLAUDE.md,
  .github/copilot-instructions.md), preserving user content byte-for-byte; specrew update refreshes it and
  specrew start heals a missing or stale section. Content comes from one packaged fragment in the module
  FileList, and the shared delivery core is host-neutral (manifest-driven, host-coupling-firewall-guarded).
  This closes the last Antigravity parity gap: a cold-launched agent comes up as the governed Specrew
  coordinator instead of a blank session. Real-host-validated for strong models (Opus 4.6, Claude).

Notes

• Weak-model caveat (FR-017): Gemini Flash facilitated the governed workshop competently but then
  self-authorized lifecycle boundaries — the coordinator must be a strong model until a deterministic gate
  (Proposal 180) lands. No full or verified Antigravity-parity claim; release carry-forwards
  (beta-before-stable, MigrateLegacyTopLevelEventMap legacy-upgrade validation, machine-local agy
  evidence) remain open.

Specrew v0.37.0

Stable promotion of the 0.37.0-beta1 line (Feature 174 — Hook-Driven Session Bootstrap), validated by the
maintainer per the beta-before-stable mandate. Builds on the 0.36.0 (Work-Kind & Branch Governance) base.

• Feature 174 — Hook-Driven Session Bootstrap (Proposal 172): automatic session orientation via host
  hooks — no specrew start required. Bundles rolling cross-session handover, approval integrity on resume,
  specrew hooks status | install | remove, specrew handover author, user-profile intake at
  specrew init, Codex double-fire hardening, and per-host lean directives that stay under the 10 K hook
  output cap.

Specrew v0.37.0-beta1

Added

• specrew handover author: writes the agent's interpretive handover notes — open questions, working
  hypothesis, and the recommended next step — into the rolling handover, so the next session inherits that
  context instead of a blank slate. It reads a Markdown packet via --from <file> or --stdin (the ##
  headers name the sections) and accepts --feature, --boundary, and --host.
• specrew hooks status | install | remove [--host <h>]: the command for installing, repairing, and
  checking your Specrew session hooks.

Changed

• Automatic session bootstrap (specrew start is now optional): after specrew init, just launch your
  host (for example, run claude or codex) and Specrew bootstraps you automatically — it greets you with an
  orientation banner and drives the governed lifecycle. You no longer have to run specrew start first;
  specrew start remains available as an explicit way to drive or re-anchor, and it is the entry point on
  Antigravity, which has no hook surface.
• Rolling handover with cross-session, cross-host auto-resume: Specrew now keeps a rolling handover of your
  work so the next session picks up where you left off. When you come back — in the same host or a different
  one — your in-progress feature follows you and the lifecycle resumes automatically. If you switch to a
  non-Claude host mid-feature, the new session may ask you to re-confirm your last approval.
• Approvals stay real on resume: when a resumed session lands at a boundary that has not been approved,
  Specrew stops and waits for your verdict. The agent will not advance on a bare continue, a single approval
  moves you forward by at most one boundary, and your approval is never invented for you.

Fixed

• Session bootstrap on Codex: the Specrew session hook now runs and surfaces its orientation on Codex,
  and the bootstrap shows the correct Specrew version.

Specrew v0.36.0

What's Changed

• docs: propose work-kind branch governance by @alonf in #2528
• chore(177): record feature closeout after 0.35.0 promotion by @alonf in #2529
• docs: refresh user docs to match 0.35.0 shipped behavior by @alonf in #2595
• docs(proposal): 184 — console experience layer + lens thought cards (renumbered from 180) by @alonf in #2597
• docs(proposals): add methodology documentation scope split by @alonf in #2598
• docs(proposal): add crew mission alias by @alonf in #2600
• Proposal 187: Volatile runtime dependency monitoring by @alonf in #2601
• [codex] add host-neutral boundary packet enforcement proposal by @alonf in #2602
• Clarify variance agility in proposal 174 by @alonf in #2603
• feat(182): Work Kind and Branch Governance Model (0.36.0-beta1) by @alonf in #2604

Full Changelog: v0.35.0...v0.36.0

Specrew v0.36.0-beta1

What's Changed

• docs: propose work-kind branch governance by @alonf in #2528
• chore(177): record feature closeout after 0.35.0 promotion by @alonf in #2529
• docs: refresh user docs to match 0.35.0 shipped behavior by @alonf in #2595
• docs(proposal): 184 — console experience layer + lens thought cards (renumbered from 180) by @alonf in #2597
• docs(proposals): add methodology documentation scope split by @alonf in #2598
• docs(proposal): add crew mission alias by @alonf in #2600
• Proposal 187: Volatile runtime dependency monitoring by @alonf in #2601
• [codex] add host-neutral boundary packet enforcement proposal by @alonf in #2602
• Clarify variance agility in proposal 174 by @alonf in #2603
• feat(182): Work Kind and Branch Governance Model (0.36.0-beta1) by @alonf in #2604

Full Changelog: v0.35.0...v0.36.0-beta1

Specrew v0.35.0

Stable promotion of the 0.35.0 line per the beta-before-stable mandate, validated by the maintainer's install testing across the 0.33–0.35 line on the Claude host (a full governed lifecycle building a cross-platform .NET MAUI app, head-to-head against an ungoverned control). Bundles the prerelease:

• Feature 177 — Code & Implementation lens (code-implementation) + specrew-code-rules guidance skill (beta1): a design-workshop lens that captures implementation craft as binding constraints and actively guides the coding agent at implement time — guidance, not a gate.

Specrew v0.35.0-beta1

Added

• Feature 177 — Code & Implementation lens (code-implementation) + specrew-code-rules guidance skill (Proposal 163). A design-workshop lens that captures how the code is written (implementation craft) as binding constraints, then actively guides the coding agent at implement time. It is conduct-driven and auto-on for any code-writing feature (like product-domain, not a row in applicability-map.json), and runs after the technical lenses so it can bind the resolved stack and architecture. The lens drives a source-of-truth-first intake (asks for an existing coding guideline or one or more example projects to emulate -- GitHub, local, or other -- before anything else, the Figma-equivalent question), an assisted ingestion that maps a provided guideline/example onto the catalog and extracts non-catalog conventions as custom rules with provenance, a grouped, pre-checked set/unset checklist (baseline summarized, decision-prompts paced all-at-once or one-at-a-time, applicability-filtered rules shown only in context -- never a flat wall), and a dependency / tooling selection decision area (FR-013) that presents "use existing project tools / no new dependency" first and, for any chosen dependency, captures version, license, source org, canonical URL, maintenance signal, security/advisory status, compatibility, cost/quota, coupling weight, replaceability, and test implications into the manifest dependency_policy. Content lives in a data-driven catalog (code-rules.yml: cross-language baseline-craft defaults plus per-stack rules for C#/.NET, C/C++, TS/JS, Python, Go, Java, including SOLID, Strategy/State over repeated conditionals, and deliberate functional-vs-inheritance polymorphism); selection lives in a per-feature reference-by-ID manifest (implementation-rules.yml, schema implementation-rules.schema.json, authored by hand following the schema like product-domain); delivery lives in one static multi-host specrew-code-rules skill that resolves the active feature, composes BASELINE plus the feature OVERLAY (plus an optional code-rules.local.yml project overlay for company/org rules), and surfaces it task-scoped while writing code. Run cadence: the rules are mostly product-level -- decide once at a product-level workshop, inherit per feature, and re-open only the parts a new technology or programming language changes (context_scope hooks; V1 feature_standalone; forward-compatible with Proposal 162). It is guidance, not a gate: there is no Proposal-145 review-time conformance engine; the acceptance proof is a deployed-module dogfood (the agent is actually guided, the human is not walled, and a new dependency is never added without surfacing the decision -- SC-004 / SC-007 / SC-008).

Specrew v0.34.0-beta1

Added

• Feature 176 — Product & Problem Domain lens (first workshop lens) (#2285). A required first design-workshop phase, product-domain, that runs before technical-lens applicability selection and grounds product/problem context (users & stakeholders, pain/job, MVP, non-goals, constraints, outcomes, alternatives) at adaptive depth (Light/Standard/Deep by risk and novelty), tagging every material statement with an evidence quality (known / assumed / unknown / research-needed). It persists a human-readable and a structured record (workshop/product-domain.{md,yml}), is enforced at the specify boundary (a batch "confirm all" cannot satisfy it), conditionally blocks plan on a load-bearing research-needed gap, and runs before every feature at adaptive depth (context_scope: feature_standalone; product_id / product_context_ref forward-compat hooks for Proposal 162). Deferred to forward-compatible shape: Proposal 156 workshop-decisions.yml emission and Proposal 162 inheritance behavior.

Specrew v0.33.0-beta2

Fixed

• Bug bash — GitHub Actions Node 24 readiness (#2214 / #2080). Updated active workflows, Squad workflow templates, and bundled GitHub templates to current Node 24 JavaScript action majors, and added a deterministic stale-pin guard.
• Bug bash — Claude-only specrew-gate-stop deployment (#2215 / #2083). Scoped the boundary verdict packet collapse fix to Claude so non-Claude hosts keep their normal rendering path.
• Bug bash — iteration state.md truth (#2216 / #2213). Kept iteration state aligned with task-progress transitions and added stale-state checks around review/recovery flows.
• Bug bash — workshop confirmation integrity (#2217 / #2212). Required scoped per-lens workshop confirmation evidence, blocked agenda/lens approval from masquerading as workshop-question approval, and tightened the dogfood gap that allowed a batch "Confirm all as proposed" shortcut to replace per-lens interaction.