agentsclimarketplace

Distill domain knowledge

Skill cat2000/domain-knowledge-ops/.cursor/skills/distill-domain-knowledge

Advanced: Compose S3→S7 or redo Recognize/S2 when materialized/ already exists (no re-sync). Default resume after wiki prep is saying continue in the same thread — use this skill for partial steps or a fresh Compose session. Full pipeline: @generate-knowledge-from-wiki.From its SKILL.md

Install
npx -y skills add cat2000/domain-knowledge-ops --skill distill-domain-knowledge

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.5 KB, 633 tokens by cl100k_base, as published. Nobody here has run it

distill-domain-knowledge (Recognize or Compose — no Ingest)

Advanced resume. After wiki prep, the default is: human confirm → say continue (same thread as @generate-knowledge-from-wiki). Use this skill when you need Compose/Recognize without re-running Ingest, or a single-step / fresh-session resume.

Playbook: generate-knowledge-from-wiki/RUNBOOK.md (three stages; steps S1–S7). Chinese locale: RUNBOOK.zh-CN.md.

confirm / continue — same meaning as wiki skill. (zh-CN strings for these tokens: domain-knowledge/language/deliverable-locale-tokens.json.)

When to @

ScenarioDo
Just finished wiki prep (same thread)Wait for confirm, then say continuedo not require @distill
Checklist already confirm; new chat / no wiki thread@distill-domain-knowledge <R> → Compose S3→S7
Re-run Recognize only@distill-domain-knowledge <R> + “S2”
One theme sub-step@distill-domain-knowledge <R> + theme <slug> S3 / S4 / S5 / S6 / S7

Agent

  1. Open RUNBOOK and domain-knowledge/distill-authoring-contract.md; run the requested range (S2: python3 scripts/distill/s2_recognize.py --root-id <R>).
  2. S3 / S4 / S5 / S6: do not translate. S6 = source brief (*-source-brief.md). S7 = locale / reader brief (*-domain-brief.md for en; zh-CN filename from deliverable-locale-tokens.json).
  3. Keep industry axes: land evidence into strategy axes per industry-axis-remount.md; do not promote source-tree layout into modules or hard-code tenant remount tables.
  4. S2 contract: require S2_DECISION_LEDGER.json + S2_REVIEW_DECISIONS.json; human entry is DOMAIN_MODULE_CHECKLIST.md. After S2: tagging_acceptance.py --root-id <R> before confirm.
  5. After S3: tagging_acceptance.py --root-id <R> --after-s3. After S7: --after-s7 --strict (zero-rule fake coverage = FAIL → revert confirm).
  6. domain_check distill: after S2 (coverage); after S7 (full).

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.