동블

그래프 뷰

README

3 min read

Repeated Work Skills

반복 작업 스킬은 같은 문제를 다시 만났을 때 원인 탐색을 처음부터 반복하지 않기 위한 짧은 실행 문서입니다. 먼저 증상과 오류 문구를 스캔해 기존 스킬과 같은 문제인지 판별하고, 같다면 해당 스킬의 해결 절차를 우선 적용합니다. 각 스킬은 작업별로 분리하며 본문은 LLM이 빠르게 읽을 수 있는 영어 체크리스트 형식으로 작성합니다.

Status: Implemented

Purpose

/docs/operations/skills/ stores reusable task runbooks for repeated repository work.

Use this directory to reduce future token cost:

  • avoid rediscovering known fixes,
  • preserve exact scan/fix/verify commands,
  • make repeated branch/worktree/documentation problems portable across dev and active work branches,
  • keep long learning docs separate from compact agent execution memory.

Required Pre-Scan

Before broad investigation, scan for a reusable skill when any symptom looks familiar.

rg -n "<symptom>|<error>|<command>|<class-or-file>" docs/operations/skills AGENTS.md docs/AGENTS.md
git status --short
git diff --name-only

Scan inputs:

  • exact error text or broken output,
  • command that failed or produced suspicious output,
  • affected file path/class/test name,
  • branch/worktree context,
  • recent changed files,
  • prior docs references.

Same-Issue Decision

Treat it as the same issue when two or more signals match:

  • same command or tool,
  • same error/symptom shape,
  • same file family or module,
  • same OS/profile/runtime context,
  • same fix verification command,
  • existing skill names the current risk.

If same issue is likely:

  1. Follow the matching skill first.
  2. Investigate only gaps not covered by the skill.
  3. Update the skill if any step was stale, missing, or misleading.

If no match:

  1. Solve normally.
  2. If this is the second occurrence of that task/problem, add a new skill doc.

File Contract

Create one file per repeated task:

/docs/operations/skills/<kebab-case-task>.md

Required structure:

# <Task Name>
 
<1-5 Korean sentences: purpose, symptoms, scope.>
 
> Status: Implemented | Proposed | Deprecated
 
## Trigger Signals
## Required Scan
## Same-Issue Criteria
## Fix Steps
## Verification
## Do Not
## Related Docs

Writing rules:

  • Body language: English or compact machine-readable checklist.
  • Prefer commands, paths, decision criteria, and concrete warnings.
  • Keep theory out unless it prevents repeated mistakes.
  • Link the new file from this README and a parent index.

Skills

Branch Portability

Skill docs must avoid branch-specific assumptions.

When this rule or a skill needs to apply to dev and unfinished branches:

  1. Keep the change docs-only.
  2. Avoid mixing it with feature, backend, frontend, or migration edits.
  3. Prefer a small commit or patch that can be merged/cherry-picked cleanly.
  4. If the current branch has unrelated uncommitted work, do not switch branches until that work is protected.
  5. After applying to another branch, run rg -n "/docs/operations/skills|Repeated-work" AGENTS.md docs/AGENTS.md docs to confirm the rule exists there.

Legacy Location

  • Historical MR-message material still exists at README.md.
  • Do not add new reusable skills there.
  • Migrate old material into task-specific files under this directory when it is touched again.

관련 문서