Changelog

CalmCraft follows Semantic Versioning. Release notes describe user-visible CLI, spec-contract, security, and compatibility changes.

0.4.3 — pending

• Merge single-chunk and finish-plan execution into run-implementation-plan, retaining explicit host-goal authorization.
• Restore Ben's review triage and implementation structure, include human and Codex feedback, and verify published fix commits before GitHub replies or thread resolution.
• Add consistent Calm Craft display icons to every skill.

0.4.2 — pending

• Limit CodeRabbit global review resolution to CodeRabbit findings and inline threads; retain the all-author merge-readiness check.

0.4.1 — pending

• Add a Calm Craft icon with a hammer badge to the plugin listing and composer.
• Expand the plugin description with supported specification, planning, review, conventions, and visualizer workflows, and starter prompts.
• Include matching portable listing metadata and verify that packaged icons are present.

0.4.0 — pending

• Restore Ben's implementation planning and execution workflows with agreed requirements and the selected plan as the delivery boundary; keep run-implementation-plan-all as a host goal loop around the runner.
• Continue work that is already authorized across skill boundaries.
• Add the reusable calmcraft dev-all development stack launcher and the shared engineering configuration contract with validation.
• Let the visualizer lead with the product contract and branch changes.

0.3.1 — 2026-09-07

• Bound browser verification to explicit checks and unresolved UI risks, reuse results across delivery phases, and distinguish setup blockers from product failures.
• Withhold branch deletion advice until stale worktree registrations are pruned, require confirmed inline resolution before global review resolution, and guard PR descriptions against unverified base refs.
• Build PR metadata from the verified published head so local commits cannot be described as part of the PR; label the regression-context example as JavaScript.
• Repair Windows registry smoke checks by invoking JavaScript entry points and resolving temporary-directory aliases before cleanup.

0.3.0 — 2026-09-06

Includes the previously unreleased 0.2.1 skill changes.

• Centralize test value and verification ownership: reuse current evidence, assign narrow worker checks, and reserve full suites for explicit checks requests, repository requirements, and release preparation.
• Keep all 31 skills while removing repeated checklists, trigger lists, and question ceremony. Add explicit code-slop cleanup and align focused browser verification with task and repository policy.
• Keep CodeRabbit explicitly invoked, account for every inline severity, retain unverified findings, and require published fixes and current thread evidence before resolution.
• Separate local readiness from authorized draft promotion; preserve PR ticket relationships and inspect the actual branch commit range.
• Accept one unguarded flow fallback, correct spec status rollups, and preserve stable IDs and generated-diagram authority.
• Require branch-cleanup PR merge evidence to reach trunk, preserve unknown Graphite ownership, and evaluate worktree blockers separately. No-fetch scans no longer write Git objects or refresh indexes.

0.2.0 — 2026-08-25

• Add spec-storyboard-journey, authoritative storyboard evidence on flow states, scene inspection in Feature view, a shared UX journey reference, and packaged spec/flow templates for reconstructing a repository's portable spec layer.
• Add calmcraft generate, which writes the estate as one self-contained HTML file that opens from the filesystem with no server, port, or token.
• Add generate --diff --base <ref> and --provenance, which bake Branch Review into that file at generation time. --provenance chooses which layers are visible when the file first opens.
• Add Flows and Questions as first-class views, and stop counting unresolved questions as findings.
• Read flow references that wrap across several lines; they previously parsed as no flow at all.
• Rebuild the interface on bundled Geist with a single type scale and reserved semantic colour.
• Render every matching specification instead of the first 120.
• run-implementation-plan is the scoped delivery loop: read the specs, build a queue, work one card at a time, and keep going across turns until the definition of done. Host /goal (Cursor, Codex, and others) starts and continues it. An optional .engineering/goal.md overlay holds repo-specific commands. The plugin no longer ships a goal skill — Cursor now has its own. Each card uses cheap, diff-scoped checks (targeted tests, lint/format on touched files, path-scoped types). Whole-programme typecheck, knip, lint, and the full test suite run once at close-out and loop until green.
• Add ask-questions: surface real open decisions in the current conversation, plan, or session-owned branch and put them to you as structured choices. Never invents the questions.
• Add branch-cleanup, update-pr, and the CodeRabbit review trio (coderabbit-review-triage, coderabbit-review-implement, coderabbit-review-implement-all). Repos without CodeRabbit simply never trigger those three.
• ready-for-pr marks a draft PR ready after the gates pass. Submit stays draft; this skill is what publishes the review.
• Ship skill sidecars in the npm package (scripts/, extra skill markdown). Select chevrons work under CSP. Muted chrome text meets contrast.

0.1.0 — pending

• Add the local Atlas, Feature, Branch Review, and Health views.
• Add semantic comparison for committed, staged, unstaged, and untracked spec changes.
• Add read-only local checkout, linked worktree, and temporary private remote sessions.
• Add the CalmCraft Agent Plugin skills and format references to the npm package.