AGENT SKILL

project-knowledge

Captures hard-won technical understanding from difficult debugging sessions and non-obvious fixes

License 许可
MIT
最近更新
2026-09-25
Stars
★ 0
Forks
0

01 前言 Front Matter

name
project-knowledge
description
Captures hard-won technical understanding from difficult debugging sessions and non-obvious fixes

03 SKILL.md 正文

Project Knowledge Skill

Captures hard-won technical understanding when agents work through genuinely difficult problems. The knowledge is preserved for future sessions.

Different from reflections: Reflections capture process failures. Project knowledge captures technical victories — the problem was genuinely hard, it was figured out, and that understanding should persist.

Triggers

  1. Stakeholder excitement — "it works!", "finally!", "that fixed it"
  2. Resolution after struggle — multiple failed attempts, now working
  3. Non-obvious fix — surprising, counterintuitive, poorly documented
  4. Agent self-recognition — unusually difficult path to solution

Process

  1. Confirm with stakeholder: "This was hard-won knowledge. Want me to record it?"

  2. Gather context: Review the conversation for the original problem, symptoms, failed attempts, working solution, and underlying explanation.

  3. Write knowledge file: Create docs/knowledge/YYYY-MM-DD-slug.md:

    ---
    date: YYYY-MM-DD
    tags: [relevant, searchable, terms]
    related-tickets: [NNN]
    ---
    

    Sections:

    • Problem — What was broken
    • Symptoms — Observable behavior
    • What Was Tried — Each attempt and why it failed
    • What Worked — The actual fix with enough detail to reproduce
    • Why It Works — The underlying explanation
    • Future Guidance — Actionable rules for future agents
  4. Commit: docs: record project knowledge -- <title>

Output

  • Knowledge file in docs/knowledge/
  • Commit with descriptive message

04 同仓库更多 Skills