AGENT SKILL
knowledge-capture
Extract durable learnings and questions after meaningful work.
- License 许可
- MIT
- 最近更新
- 2026-08-08
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- knowledge-capture
- description
- Extract durable learnings and questions after meaningful work.
03 SKILL.md 正文
Knowledge Capture
Purpose
Separate active task state from durable knowledge proposals.
Extract
- decisions made
- assumptions introduced
- assumptions disproved
- new domain relationships
- undocumented behaviours
- integration quirks
- failure modes
- reusable patterns
- contract changes
- operational risks
- unanswered questions
- contradictions with existing guidance
Destination Decision
active resume state -> HANDOFF.toon
unanswered material issue -> knowledge/questions
evidence-backed discovery -> knowledge/learnings or inbox
durable decision -> decision proposal
reusable practice -> pattern proposal
failure mode -> risk proposal
trivial implementation -> no durable record
Rules
- Do not write directly to canonical knowledge unless reviewed evidence already exists or the user explicitly directs promotion.
- Prefer
.agents/knowledge/inbox/for unreviewed proposals. - Link proposals to existing IDs when possible.
- Avoid recording trivial implementation details.
- Capture contradictions with existing guidance.
- Update
HANDOFF.toon.knowledgeon every meaningful task: record consulted IDs/paths, proposals created, or a concreteno_recordreason.
Output
handoff_updates: []
knowledge_proposals:
- path: .agents/knowledge/inbox/INBOX-001-example.md
reason: possible reusable pattern
questions_created: []
no_record:
- reason: local-only mechanical change
04 同仓库更多 Skills
agentic-template
Start or continue work in this repository using the canonical AGENTS.md contract and repository skill catalog.
★ 1
wiki-tidy
Reconcile the wiki against the knowledge graph, specs and code; prune stale, keep links resolving.
★ 1
update-handoff
Update HANDOFF.toon with current semantic work state.
★ 1
test-first
Drive meaningful behaviour changes with the smallest useful failing test.
★ 1
test-change
Select and run tests that prove a change at suitable fidelity.
★ 1
specify
Create proportional OpenSpec-shaped structured change proposals for meaningful changes.
★ 1