v0.8.0

• Reduced unnecessary PPP pauses by letting clear single-task work auto-advance until a real decision, expensive proof step, or blocker appears.
• Tightened Shape so it still confirms one decision at a time, but only for material trade-offs instead of low-risk defaults and obvious routing.
• Simplified Ship's visible output so it always reports route and confidence, but only shows the full routing scorecard when it helps the user steer.
• Upgraded Ship's execution-wave planning to detect foundation-first work more aggressively, block unsafe parallelism around unstable shared contracts, prefer two-wave foundation-then-parallel plans, and reason explicitly about review burden.
• Strengthened PPP and PPP Cloud proof heuristics to choose behaviour-direct proof first, reject broad but weak validation bundles, and make the primary proof more visible in review and PR handoff.
• Improved PPP and PPP Cloud blocker handoffs with explicit proof gaps and smallest next resumable tasks after failed validation attempts.
• Improved Shape stop and handoff quality so blocked shaping reports what is already safe, what remains intentionally unshaped, and the smallest next shaped task after the missing decision.
• Added a stronger minimal-output bias across the core skills while preserving important assumptions, proof choices, risks, blockers, and routing rationale.
• Added a skill eval harness with repo-local prompts, granular weighted criteria, penalties, dimension-level scoring, run scaffolding, result fingerprints, baseline comparison, a cross-model scoreboard, adversarial cases, and CI validation of both the eval definitions and the workflow itself.
• Added generated manual-review.md companion sheets for eval runs so human reviewers can capture best-route, best-proof, blocker-quality, verbosity, and rubric-calibration judgments alongside machine scores.
• Added structured reasoning_quality_notes to every eval case so human reviewers can calibrate responses against explicit weak, good, and excellent answer patterns.

v0.7.0

• Added shape and ship as first-class shipped skills in the installers and uninstallers.
• Versioned all shipped skills consistently and updated release automation to validate and bump all four skills.
• Updated README, setup docs, troubleshooting, and templates to explain when to use shape, ship, /ppp, and /ppp-cloud.
• Updated shape to confirm decisions and routing recommendations interactively one at a time before handing off work.
• Added a ready-to-ship handoff flow in shape that offers spec-file creation before /ship, then re-offers /ship after the spec step.
• Tightened ship so it recommends the most efficient safe dependency-aware execution waves instead of defaulting toward overly sequential delivery.

v0.6.0

• Improved README intro and surfaced the setup decision table earlier.
• Added skills.sh quick-start to README.
• Added npm keywords to package.json for registry discoverability.
• Added skill preview GIF to README.
• Added scripts/bump_version.py to automate version bumps across package.json, skill frontmatters, and CHANGELOG.md.

v0.5.0

• Added npm trusted-publishing support to the release workflow.
• Added release validation for package.json version matching the release tag.
• Added npm package metadata fields for repository, homepage, and issue tracker URLs.
• Added .tgz package artifacts to .gitignore.

v0.4.0

• Added a native npx installer via bin/ai-engineering-starter-kit.js.
• Added package.json for npm distribution and CLI aliases.
• Added support for install, uninstall, --dry-run, --force, --yes, and --repo-local in the Node CLI.
• Updated README install guidance to prefer npx ai-engineering-starter-kit install.

v0.3.0

This release improves onboarding, install safety, and release-readiness for PPP and ppp-cloud.

• Added README badges for CI, license, and GitHub releases.
• Added top-level "How it works" diagrams for /ppp and ppp-cloud.
• Added a README setup decision table covering personal install, repo rollout, Cursor, and cloud-agent usage.
• Added README positioning on how PPP differs from Matt Pocock's skills and gstack.
• Clarified slash-command support and promoted the always-works fallback invocation.
• Added explicit repo-local skill install commands for .github/skills.
• Added a concrete README bug-fix example that links to the full PPP transcript.
• Added scripts/validate_skills.py and a CI check for skill frontmatter and unclosed code fences.
• Added installer safety features: --dry-run, --force, backups, version output, and overwrite prompts.
• Added a .github/ISSUE_TEMPLATE/ppp-cloud-task.md template for assigning bounded cloud-agent tasks.
• Added a release automation spec recommending tag-driven GitHub Releases.