Knowledge Run

Drive the inverted pipeline end-to-end for one fresh topic, in a single invocation. An operator researching a new topic otherwise hand-dispatches ~7 phase skills (knowledge-plan → knowledge-curate → knowledge-fetch → knowledge-ingest → knowledge-distill → knowledge-compose → knowledge-verify → knowledge-finalize), re-reading each SKILL.md and threading the project path between them. knowledge-run is the ordered-phase driver over that chain: one --topic, all phases in order, ending with a deposited type: synthesis page in the bound wiki.

This skill is a pure orchestrator — it composes existing cogni-knowledge phase skills via Skill(...), never re-implementing them. It dispatches zero cogni-research skills and runs no vendored wiki-* script directly; ingest fan-out, per-batch merge, and the integrity sweep — plus compose expansion and the accept check — come "for free" because knowledge-ingest and knowledge-compose already encapsulate them.

knowledge-run is the fresh-topic parallel entrypoint to knowledge-refresh --mode push: refresh lints the bound wiki and re-researches stale pages selected interactively; knowledge-run takes a single explicit topic and runs one chain. The Phases-2–7 dispatch block is identical between the two — this skill mirrors that proven loop rather than writing new orchestration, so the two entrypoints cannot drift.

Read ${CLAUDE_PLUGIN_ROOT}/references/delegation-contract.md once per session to remember the delegation boundary and the Skill(...)-dispatch convention (§"How Skill(...) blocks are written").

When to run

• The user wants to research a brand-new topic into a bound knowledge base without dispatching each phase by hand.
• The user has a sharp topic and wants the whole plan → finalize chain driven autonomously to a deposited synthesis.

Never run when