.agents/skills/deep-review/references/dimensions/skill-freshness.md
Agent skills are load-bearing documentation: agents follow them literally. A diff that changes behavior a skill describes silently poisons every future agent session. This dimension keeps the skill library in sync with reality and spots knowledge worth capturing. Findings are advisories — they skip the verify pass.
AGENTS.md / CLAUDE.md / README setup docs?rg each old name across .agents/skills/ and root-level agent docs (AGENTS.md, CLAUDE.md) — a hit on a removed/renamed thing is a stale-skill finding.