agent-rules 規則包：Conventional Commit + 6 視角 PR 審查 + 5 Whys 根因分析

# Agent Rule Pack — pick ONE mode per invocation
# Works as a Claude Code / Cursor slash command. Set {{MODE}} to: commit | pr-review | five-whys

MODE: {{MODE}}
CONTEXT: {{CONTEXT}}   # e.g. the diff, the PR link, or the problem statement

=== MODE: commit ===
Create well-formatted, ATOMIC commits.
1. Run `git status`. If nothing is staged, review changes and stage the relevant ones.
2. Run pre-commit checks if the project has them (lint, build, type-check). Fix failures before committing.
3. If the staged changes cover unrelated concerns, SPLIT them into multiple commits — one concern each.
4. Pick the type and emoji, then write `type(scope): description` (imperative, lowercase, no trailing period). Add a body explaining WHY for non-trivial changes.
   ✨ feat  | 🐛 fix | 📝 docs | ♻️ refactor | 🎨 style | ⚡️ perf | ✅ test | 🧑‍💻 chore | 🚧 wip | 🔥 remove | 🚑 hotfix | 🔒 security

=== MODE: pr-review ===
Review the change through SIX personas, each with its own checklist. Report findings per persona.
1. Product Manager — does it deliver the intended business value and meet requirements?
2. Developer — maintainability, standards, performance, scalability.
3. Quality Engineer — test coverage, edge cases, regression risk.
4. Security Engineer — vulnerabilities, authn/authz, input validation, data protection.
5. DevOps — CI/CD, config, monitoring, rollback safety.
6. UI/UX Designer — visual consistency, usability, accessibility, responsiveness.
RULE: anything you'd "fix later" must be raised NOW. End with: APPROVE or REQUEST CHANGES + the blocking items.

=== MODE: five-whys ===
Find the ROOT cause, not the symptom. Focus on process, not people.
1. State the problem/symptom precisely.
2. Ask "why" five times in a chain; each answer feeds the next question (Why 1 → Why 5).
3. Validate the chain: would fixing the final cause actually prevent recurrence?
4. Propose solutions that target the ROOT cause, plus a preventive measure so it can't return.
Output: the numbered Why-chain, a one-line root-cause statement, then the fix + prevention.

# Agent Rule Pack — pick ONE mode per invocation
# Works as a Claude Code / Cursor slash command. Set {{MODE}} to: commit | pr-review | five-whys

MODE: {{MODE}}
CONTEXT: {{CONTEXT}}   # e.g. the diff, the PR link, or the problem statement

=== MODE: commit ===
Create well-formatted, ATOMIC commits.
1. Run `git status`. If nothing is staged, review changes and stage the relevant ones.
2. Run pre-commit checks if the project has them (lint, build, type-check). Fix failures before committing.
3. If the staged changes cover unrelated concerns, SPLIT them into multiple commits — one concern each.
4. Pick the type and emoji, then write `type(scope): description` (imperative, lowercase, no trailing period). Add a body explaining WHY for non-trivial changes.
   ✨ feat  | 🐛 fix | 📝 docs | ♻️ refactor | 🎨 style | ⚡️ perf | ✅ test | 🧑‍💻 chore | 🚧 wip | 🔥 remove | 🚑 hotfix | 🔒 security

=== MODE: pr-review ===
Review the change through SIX personas, each with its own checklist. Report findings per persona.
1. Product Manager — does it deliver the intended business value and meet requirements?
2. Developer — maintainability, standards, performance, scalability.
3. Quality Engineer — test coverage, edge cases, regression risk.
4. Security Engineer — vulnerabilities, authn/authz, input validation, data protection.
5. DevOps — CI/CD, config, monitoring, rollback safety.
6. UI/UX Designer — visual consistency, usability, accessibility, responsiveness.
RULE: anything you'd "fix later" must be raised NOW. End with: APPROVE or REQUEST CHANGES + the blocking items.

=== MODE: five-whys ===
Find the ROOT cause, not the symptom. Focus on process, not people.
1. State the problem/symptom precisely.
2. Ask "why" five times in a chain; each answer feeds the next question (Why 1 → Why 5).
3. Validate the chain: would fixing the final cause actually prevent recurrence?
4. Propose solutions that target the ROOT cause, plus a preventive measure so it can't return.
Output: the numbered Why-chain, a one-line root-cause statement, then the fix + prevention.