wiki-refresh
Wiki Refresh
Close the update loop in the cogni-research → cogni-wiki integration. Wiki pages age (wiki-lint flags stale_page >365d, stale_draft >180d), but lint alone doesn't bring fresh evidence. This skill matches stale pages to sub-questions of a completed cogni-research project, materialises one synthesis file per matched pair, and pipes each through wiki-update to apply the diff against the existing page.
This is a pull-mode primitive — the user provides the fresh research project; this skill never spins up new research. (Push-mode auto-research per stale page is deferred — it would cost ~$0.50/page and surprises a wiki owner with a research bill.)
The skill is a pure orchestrator. It calls lint_wiki.py directly for staleness data (cheaper than dispatching wiki-lint, which would write a noise line to wiki/log.md), runs refresh_planner.py for the match plan, materialises per-page refresh files under <wiki-root>/raw/refresh-<research-slug>-<YYYY-MM-DD>/, and then dispatches wiki-update sequentially per matched page. Every wiki write goes through existing locked code paths.
Read ${CLAUDE_PLUGIN_ROOT}/references/karpathy-pattern.md once before proceeding to re-anchor on the three-layer model (raw / wiki / schema) — refresh content lands in raw/, not directly in the per-type page dirs.
When to run
- User asks to refresh stale pages from a specific completed cogni-research project
- User wants to pull fresh evidence into an aging wiki without re-running research from scratch
- After a periodic deep-research run on a known domain, to fan the new findings into existing wiki pages