Skill evolution validator

Skill Evolution Validator

Use this skill for manual, report-first checks of the installed evolution framework. It gathers evidence and reports risks; it does not edit other skills unless the user separately requests a repair.

When the same request explicitly asks to fix, repair, update, or resolve findings, complete the report first and emit a repair handoff owned by skill-evolution-core and skill-evolution-router. The validator still does not edit files. The owning workflow applies the authorized change, updates local ledgers, and returns for a full rerun without asking for duplicate approval.

Trigger Boundary

Use this skill when the user explicitly asks for an evolution-system:

• self-check, health check, cleanliness check, or audit
• trigger or routing behavior regression
• local-ledger comparison
• platform freshness review
• release or GitHub readiness report
• global/project rule-authority and instruction-loading-budget review
• installed versus public-framework source comparison

Do not trigger it for routine skill edits, configured evolution shortcuts, configured absorption shortcuts, or ordinary rule-maintenance requests. Those belong to skill-evolution-core and skill-evolution-router.

Modes

• auto: run full when no previous report or snapshot exists; otherwise run log-based fast mode.
• full: run structure, semantics, behavior regression, event evidence, ledger reconciliation, rule authority, instruction-loading budget, framework-source comparison, and platform freshness.
• fast: compare current files with the previous snapshot and summarize existing event evidence. It must say that behavior regression was not run.

Run:

python scripts/validate_evolution_skills.py --mode auto

Use --mode full before public release or after routing behavior changes. Use --project-root for the active repository, --framework-root for a public checkout, and --repair-authorized only when the current request already authorizes remediation.

Required Output

Every run writes Markdown, JSON, a latest snapshot, and timestamped snapshot history under the local report directory. The report must distinguish Hook definition/trust health, real observations, and smoke-test evidence; list unlabeled route outcomes; show files not represented by the latest local ledger entry; report authority conflicts, global guidance loading budget, and managed-source drift; and state whether the official platform review is stale.

References

• references/check-matrix.md: mode-specific checks and boundaries.
• references/report-schema.md: report fields and wording requirements.
• references/ledger-reconciliation.md: local change-log format and matching rules.
• references/devolution-self-check-boundary.md: validator versus rule-maintenance ownership.
• references/regression-cases.md: human-readable behavior expectations.
• references/regression-cases.json: executable behavior cases.
• references/severity-levels.md: finding severity definitions.
• references/platform-freshness.json: official-source review timestamp.