meta-docs
Meta Docs
Objective
- Separate metadata search from body reads to minimize context usage.
- Keep the docs knowledge base consistent with structured updates and logs.
Directory and Metadata Rules
- All documents live under
docs/. Directory structure is not enforced. - Avoid
docs/{TYPE}as a convention; do not create directories purely by type. - Classification is defined by the
typemetadata (single string). - Recommended types:
design,spec,guide,log,reference,decision,research,meeting,incident,runbook,roadmap,report,checklist,retro,note
Every document must include YAML frontmatter:
---
title: "Document title"
created: YYYY-MM-DD
updated: YYYY-MM-DD
author: "name <email>"
editors: ["name <email>"]
type: "design"
tags: ["ai", "troubleshooting", "gitlab"]
history:
- "YYYY-MM-DD name <email>: initial entry"
- "YYYY-MM-DD name <email>: change summary"
---
author/editorsshould use Git user info (user.name,user.email). If unavailable, fall back to the system username.
Prerequisites
This skill assumes uv for execution.
# Install uv if needed
curl -LsSf https://astral.sh/uv/install.sh | sh
Recommended execution style:
uv run --project skills/meta-docs skills/meta-docs/doc_manager.py <command> [options]
Use --root when the project root differs from the current working directory. You can place --root before or after the subcommand.
Mandatory Behavior Rules
- Context discipline: never use
cat,grep, orfindto locate docs; always runsearchfirst. - Selective reads: choose the minimal set of files from search results and read only those.
- Active maintenance: after code or architecture changes, always call
update. - Error capitalization: after resolving complex bugs, create a troubleshooting log with
create.
Commands
1) search
- Input:
--tags "keyword",--type "design",--dir "path"(optional) - Behavior: parse only YAML frontmatter and output matching docs as a JSON array
Example:
uv run --project skills/meta-docs skills/meta-docs/doc_manager.py search --tags "ai troubleshooting" --type "incident" --dir "team"
2) read
- Input:
--path "docs/target.md" - Behavior: return only the document body (frontmatter excluded)
Example:
uv run --project skills/meta-docs skills/meta-docs/doc_manager.py read --path "docs/20260312-meta-docs-intro.md"
3) update
- Input:
--path "..." --log "change summary" - Behavior: update
updated, appendhistoryasYYYY-MM-DD name <email>: log - If
authoris missing, it is auto-filled with the current editor - The current editor is appended to
editors
Example:
uv run --project skills/meta-docs skills/meta-docs/doc_manager.py update --path "docs/20260312-meta-docs-intro.md" --log "architecture updates"
4) create
- Input:
--title "..." --tags "..." --content "..." [--type "log"] - Behavior: create
YYYYMMDD-{TITLE}.mdunderdocs/(type defaults tolog, override with--type) - Title rule: use English titles so filename slugs stay within lowercase a-z, digits,
-, and_ author/editorsprefer Git user info and fall back to system username
Example:
uv run --project skills/meta-docs skills/meta-docs/doc_manager.py create --title "redis-timeout" --tags "infra troubleshooting" --content "Root cause and fix" --type "incident"
More from dhkimxx/ai-agent-skills
datasheet-intelligence
This skill is triggered when a request needs datasheet/TRM-grounded hardware facts or firmware init code with page/section citations. Keywords: datasheet, TRM, register map, base address, bitfield, reset value, pin mux, clock divider, init code, 데이터시트, 레지스터, 초기화 코드.
24naver-land-scouter
네이버 부동산 데이터 탐색과 분석을 위한 스킬. 지도 기반 단지 탐색, 단지 심층 분석, 매물 비교, 투자 지표 산출, 학군/교통 요약 리포트 생성에 사용한다. Keywords: 네이버 부동산, 매물 찾기, 시세 분석, 단지 정보, 부동산 임장 보고서, naver land, complex, article, price, school, transportation
8librarian-tool
A frontmatter-first docs workflow that separates metadata search from body reads. Keywords: docs, frontmatter, metadata, search, librarian.
1