freshness-sweeplisted

# Freshness sweep A knowledge pack rots silently: a skill pinned to "Ktor 3.x" or "Next.js 16" keeps asserting stale facts with full confidence long after the ecosystem moves. Stale knowledge is worse than no knowledge - the model would otherwise hedge or search. ## Procedure 1. **Pick targets** (max 10 per sweep, rotate): the most version-fragile skills first. Fragility ranking: - pinned major versions in the name or first paragraph (nextjs-turbopack, swift-concurrency-6-2, ios26-liquid-glass, angular-developer, bun-runtime, kotlin-exposed, kotlin-ktor, nuxt4-patterns, react-performance, remotion) - hosted API endpoints and pricing (fal-ai-media, nutrient-api, mailtrap-email-integration, free-tier-scraper-apis, x-api, agent-payment-x402) - stable references (regex-llm-hybrid, ffmpeg-media-recipes, wcag22-reference) - lowest priority, check yearly at most. Track rotation state in `.claude/skills/freshness-sweep/last-sweep.md` (date + skills checked). Additionally, every sweep: check obra/superpowers releases (source of the 12 adapted fable-workflows skills) for substantive methodology changes since the last sweep - plumbing/platform changes there do not count, only changed rules or tables. 2. **Verify each target** with web search: current stable version, breaking changes since the version the skill documents, deprecated flags/endpoints the skill still recommends. Delegate per-skill checks to subagents on sonnet - this is m