Iabstergo1

ingest

End-to-end add a new external source (PDF/DOCX/PPTX/Markdown) to the study knowledge base — deterministic preprocessing → read the whole source and write status:proposed pages + concept resolution → finish with lint and publish. Use when the user says "add this book/PDF to the KB / ingest <source> / index this document / weave this file into the wiki". Only for ingesting a new external source; read-only requests like "summarize this / explain this / translate this / answer a trivia question" must never trigger it.

kb-postmortem

Post-publish retrospective on one ingested source — collect deterministic proxy metrics (ingest-stats), refresh the failure-signal backlog with increment context (snapshot the old backlog.yaml BEFORE skill-mine, then diff), read the digest routing deviations, and write one standardized report with recommendations only (skill-evolve triggers, gate/truth updates, proposals-resolve dry-run commands). Use when the user says "postmortem this ingest / 复盘这次入库 / how did this book's ingest go / ingest retrospective for <source>". Only for a source already published; "run a KB QA / audit coverage" is kb-qa, "semantic health check" is wiki-lint-semantic, and "distill this failure into the skill" is skill-evolve — none of those trigger this.

kb-qa

Run QA / audit / coverage checks on the published knowledge base or pre-save candidates, producing a report and Review-Queue proposals. Use when the user says "run a KB QA / audit coverage / spot-check evidence / run the Q-chain / check for concept pollution". Semantic-health words (L4, contradiction, Q2 added value) belong to wiki-lint-semantic and must not be hijacked here.

kb-query

Read-only query of the existing study knowledge base, persisting a query-session (without writing any vault file). Use when the user asks "what does the KB say about X / search my wiki / what was that Y I learned / has Z ever been covered". Read-only; to keep a conclusion in the wiki, follow up explicitly with kb-save.

Code & Development Listed

kb-review

Work the Review-Queue and review_proposals items one by one (lint failure lists, cross-domain promotion candidates, overwrite-protection-rejected change proposals), giving analysis and fix suggestions; the final accept/reject is the user's. Use when the user says "work the review queue / look at the pending items / what's in the Review-Queue / walk me through the review backlog".

kb-save

Save a synthesis/comparison/learning-path/self-test candidate from an existing query-session into the wiki as status:proposed (gated, two-phase publish). Use when, after a query, the user says "save that comparison/conclusion into the wiki / form a synthesis / save to the KB / keep this as a note". One-off facts, plain explanations, and restating an existing page are not saved.

pipeline-doctor

Diagnose and safely repair a stuck ingest pipeline using only whitelisted CLI commands — stale vault locks, crashed running stages, corrupt window-done JSON, a forward-only state machine that refuses to re-run preprocessing. Use when the user says "the pipeline is stuck / 状态机卡住了 / the lock won't release / window-done keeps failing / it won't let me re-run profile / diagnose the pipeline". Never for content quality (kb-qa / wiki-lint-semantic), ingesting a source (ingest), post-publish retrospectives (kb-postmortem), or editing skills (skill-evolve); it never hand-writes SQL or edits the SQLite file directly.

skill-evolve

Distill a recurring ingest/lint failure into a bounded improvement to one skill — read the skill-mine backlog → write a bounded SKILL.md edit on an isolated branch → run skill-gate (pytest + dual-tree parity + gate-integrity) → skill-stage the candidate, leaving skill-adopt to a human. Use when the user says "distill this failure into the skill / evolve a skill / let a skill self-improve / handle skill backlog item N". Only for improving this project's own skills; "summarize this / explain this / translate this / add this book to the KB (that is ingest)" must never trigger it.

source-preflight

Run the deterministic preprocessing chain on a new external source and accept its staging artifacts, without writing any semantic wiki pages. Use when the user says "preprocess this PDF first / run source-preflight / build the source profile first / see if it can be ingested". Only the zero-LLM acceptance gate for add-source, profile, source-convert, source-audit, windows, workorder — no book-splitting, summarizing, semantic unit planning, or vault writes.

source-xray

Generate a book-breakdown reading note or synthesis-candidate report from published source/concept/topic content, writing only pipeline-workspace/reports/source-xray/ by default (not the vault). Use when the user says "x-ray this published source / book-breakdown reading notes / source-xray / generate learning-note candidates". It does not preprocess, does not decide windows, does not decide write scope, and does not create or merge concept pages.

Code & Development Listed

wiki-lint-semantic

Run the judgement-requiring health checks on the KB — whether a comparison truly covers the key difference dimensions (L4), whether cross-page conclusions contradict, whether recent kb-save output adds real value (Q2) — emitting Review-Queue proposals only, never editing any wiki page. Use when the user says "do a semantic health check / check for contradictions / see whether the comparison pages are complete". Deterministic lint (L1/L2/L3/L5/L6/broken-links/duplicates) is handled by scripts/pipeline.py lint, not here.