AGENTS.md Starter Kit v1.4.2

What changed

• scan-agent-instructions is now manual-first and self-contained: it does not download, install, or execute a remote scanner.
• An already-installed repo-agent-scan v1.2.0 or newer may supplement the review, but absence or an older version falls back to direct local inspection without automatic installation or upgrade.
• The README now recommends explicit Homebrew or Scoop installation and retains the provenance-backed v1.2.0 release digest.

Verification

• Skill Creator structural validation passed.
• All 10 repository checks passed locally and in GitHub Self-test run 33605533334 at commit 3fc71bad598518c6d7a0d17cd7085e1e68d47041.
• The package preview contains only the expected 10 files.
• The GitHub Action implementation did not change, so the reviewed mutable v1 Action alias remains unchanged.

AGENTS.md Starter Kit v1.4.1

What changed

• The bundled scan-agent-instructions skill now discovers every nested SKILL.md in addition to repository agent-instruction files.
• The skill downloads the provenance-backed repo-agent-scan v1.2.0 release and verifies SHA-256 d47c96668525357f60d0b9528118bc668f8eaabf7e2fd0fe87809536115851da before use.
• Documentation and release-drift tests now match the v1.2.0 scanner contract.

Verification

• All 10 starter-kit checks pass on commit d1542ec3eaed307677460b7787dfed34ce2dd576.
• GitHub Self-test run 33603397232 passed.
• The starter kit's GitHub Action implementation did not change, so the mutable v1 Action alias remains on the last reviewed Action release.

AGENTS.md Starter Kit v1.4.0

What changed

• Added the installable scan-agent-instructions skill and repository security workflow guidance.
• Updated the skill and README to use the provenance-backed repo-agent-instruction-security-scan v1.1.3 release asset.
• Preserved first-use installation confirmation and documented the verified SHA-256 digest.
• Added an offline regression test that keeps README and skill release references in sync.
• Added the GitHub CLI companion workflow, repository-prefilled self-audit, and one-click template path introduced since v1.3.0.

Verification

• 10 deterministic tests pass.
• JavaScript syntax checks pass.
• npm pack --dry-run contains only the expected CLI, templates, skills, README, license, and package metadata.
• GitHub Actions self-test run 33601207313 passed on the release commit.

The GitHub Action implementation is unchanged; the existing v1 Action alias remains at its last reviewed Action release.

AGENTS.md Starter Kit v1.3.0

Installable create-agentsmd skill

This release adds a portable Agent Skill that creates, revises, or audits AGENTS.md from repository evidence.

npx skills add sunxiayi/agents-md-starter-kit --skill create-agentsmd

The skill preserves accurate maintainer rules, verifies commands against manifests and CI, handles nested monorepo scope, and avoids guessed or unsafe instructions. It works with Agent Skills-compatible tools including Codex, Claude Code, Cursor, GitHub Copilot, Gemini CLI, and Windsurf.

For an optional browser-local second opinion, use the free Repo Agent Kit checker. No signup is required and pasted instruction text is not uploaded.

Verification

• all 9 repository tests pass
• Skill Creator validation passes
• npx skills discovers and installs the skill
• npm package dry run includes skills/create-agentsmd/SKILL.md

AGENTS.md Starter Kit v1.2.0

Adds a dependency-free AGENTS.md Readiness Check GitHub Action. It uses the same deterministic 100-point categories as Repo Agent Kit, writes a job summary, exposes score and grade outputs, rejects paths outside the workspace, and can run in report-only mode or enforce a configurable threshold. Use: sunxiayi/agents-md-starter-kit@v1.

AGENTS.md Starter Kit v1.1.0

Adds a dependency-free CLI that copies minimal, monorepo, Python, or Next.js templates with one GitHub-backed npx command. The CLI refuses to overwrite an existing file unless --force is explicit, supports dry runs, and performs no telemetry.

AGENTS.md Starter Kit v1.0.0

Four copy-ready AGENTS.md templates, a concise Claude Code bridge, and a repository review checklist. Every template is intentionally small, editable, and built around exact commands and observable completion criteria.