원클릭으로
bugfix
Reproduce a bug locally, trace intent and history, propose the smallest safe fix, then implement and verify the repair.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Reproduce a bug locally, trace intent and history, propose the smallest safe fix, then implement and verify the repair.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | bugfix |
| description | Reproduce a bug locally, trace intent and history, propose the smallest safe fix, then implement and verify the repair. |
Reproduce the bug first, inspect the code's intent and history, propose the smallest defensible fix before coding, then implement, verify, and optionally follow up with docs.
{{rnd/agents/developer.md}} {{rnd/agents/e2e-engineer.md}} {{rnd/agents/shared/command-hygiene.md}}
Restore broken behavior with the narrowest safe change. Start with local reproduction, trace the execution path and commit history, separate intended behavior from the regression, present a minimal fix proposal to the user, and only then implement and verify the approved repair.
This workflow is for regressions, not redesigns. The job is to understand why the current code exists, confirm the failure path, and apply the smallest change that restores the intended behavior.
git log, git blame, and relevant diffs to understand why the broken behavior was introduced.rnd/agents/developer.md and the applicable AGENTS.md / CLAUDE.md files.Ask yourself before changing code: Do I understand the failure, the original intent, and the narrowest safe fix?
Before starting any work, create a persistent phase checklist using the environment's task/todo tool or by writing a tracking file to /tmp/bugfix-<bug-id>-phases.md.
Required checklist:
- [ ] Phase 1: Reproduction gate (required skill: `run-manual-qa-tests`; fallback: `rnd/agents/manual-qa-tester.md`)
- [ ] Phase 2: Investigation (separate context required; no phase skill exists; use `rnd/agents/developer.md`)
- [ ] Phase 3: Plan review (separate context required; no phase skill exists; use `rnd/agents/developer.md`)
- [ ] Phase 4: Implementation (required skill: `implement-build-plan` when a relevant plan exists; otherwise no phase skill exists, use `rnd/agents/developer.md`)
- [ ] Phase 5: Verification (required skill: `run-manual-qa-tests`; fallback: `rnd/agents/manual-qa-tester.md`)
- [ ] Phase 6: Documentation follow-up (separate context required; no phase skill exists; use `rnd/agents/developer.md`)
- [ ] Wrap-up: Interaction log written
Rules:
Required skill: run-manual-qa-tests
Fallback teammate brief: rnd/agents/manual-qa-tester.md
Separate context required.
No phase-specific skill exists.
Use agent brief: rnd/agents/developer.md
AGENTS.md / CLAUDE.md files before editing.rnd/tech_specs/ or rnd/build_plans/ contain relevant context, read the matching files and separate intended behavior from accidental behavior.Separate context required.
No phase-specific skill exists.
Use agent brief: rnd/agents/developer.md
When a relevant build plan exists:
implement-build-planrnd/agents/developer.mdWhen no relevant build plan exists:
rnd/agents/developer.mdRequired skill: run-manual-qa-tests
Fallback teammate brief: rnd/agents/manual-qa-tester.md
Separate context required.
No phase-specific skill exists.
Use agent brief: rnd/agents/developer.md
| Input | Location | Purpose |
|---|---|---|
| Bug report | User prompt, issue link, stack trace, or observed failure | Starting point for reproduction |
| Codebase | Current repository | Failure path and integration points |
| Git history | Current repository history | Intent behind the regression |
| Repo instructions | applicable AGENTS.md / CLAUDE.md files | Stack-specific conventions and test rules |
| Specs and plans | rnd/tech_specs/ and rnd/build_plans/ when present | Intended behavior and scope boundaries |
rnd/manual-qa-results/When this skill is used directly in an editor or coding session:
{{rnd/agents/summary.md}}
Execute live manual QA from written cases or an inline repro case, capture verifiable artifacts, and write a structured result report.
Convert a manual QA Markdown result into a standalone HTML evidence report with clickable proof links for JSON, images, and code snippets.
Turn a technical spec or concrete problem statement into a concrete implementation and test plan formed of small, traceable tasks.
Review PR discussions to identify improvements to agents, templates, or instructions; write a retro report to rnd/retros.
Convert product specs into a repo-grounded technical specification / high-level design.
Produce E2E sanity test cases (English) for features and save them under rnd/test_cases/.