Skill Sunset v0.2.0

Skill Sunset v0.2.0

Skill Sunset audits Agent configuration for instructions that may be duplicated, stale, over-broad, or no longer useful as models and runtimes evolve. The core is deterministic, read-only, dependency-free, and does not call an AI service.

Highlights

• Audit Codex with npx skill-sunset@latest audit --codex --open.
• Audit Claude Code with npx skill-sunset@latest audit --claude --open.
• Produce portable English and Chinese HTML, Markdown, JSON, evaluation-plan, and Agent handoff artifacts.
• Distinguish KEEP, MERGE, UPDATE, DEMOTE, RETIRE, and unproven TEST findings.
• Redact user-facing absolute paths and keep optional experiment commands on a minimal environment by default.
• Run CI across Ubuntu, macOS, and Windows with Node.js 20, 22, and 24, plus Gitleaks secret scanning.

The animation is generated from the repository fixture report: 5 files scanned and 11 findings, including duplicate retirement, stale-reference updates, and hypotheses that still require validation.

Try it

npx skill-sunset@latest audit --codex --open

The command scans the default Codex configuration boundary, writes a portable report bundle, and opens the localized HTML report. No global install is required.

Compatibility

• Node.js 20, 22, and 24
• Ubuntu, macOS, and Windows
• Codex, Claude Code, and bounded custom configuration directories

Safety boundary

Skill Sunset reports evidence and proposed actions; it does not automatically delete or rewrite configuration. A TEST verdict is a hypothesis that must be validated against behavior before cleanup. Optional experiment commands run only when explicitly requested.

See the changelog for the complete list of changes.