claude-starter-kit

Dùng khi user nói 'api audit', 'contract check', 'kiểm tra API', 'endpoint match docs?', 'breaking change?', hoặc trước release. Verify API endpoints match api-contract.md /100.

Dùng khi user nói 'audit', 'quality check', 'kiểm tra chất lượng', 'điểm chất lượng', 'ready to merge?', 'persona coverage', hoặc trước khi merge một wave/feature lớn vào main. Chấm điểm 11 categories /110 điểm (10 tech + 1 persona coverage).

Dùng khi user nói 'business audit', 'logic check', 'kiểm tra business logic', 'code đúng rules chưa', hoặc trước GA release. Verify code implement đúng rules.md + use-cases.md. 5 categories /100.

Dùng khi user nói 'design pattern audit', 'pattern check', 'kiểm tra design patterns', 'God service đâu', 'anti-pattern hotspot', hoặc trước một refactor planning cycle. Score code against `.claude/rules/design-patterns.md` §3 BANNED list. Output: hotspot list (file path + LOC + violated pattern) + score /100.

Dùng khi user nói 'ops audit', 'production ready?', 'kiểm tra ops', 'deploy checklist', 'monitoring check', hoặc trước GA release. Production operations readiness /100.

Dùng khi user nói 'perf audit', 'performance check', 'load test', 'kiểm tra hiệu năng', 'N+1', 'bundle size', hoặc trước production deploy. Baseline performance metrics /100.

Dùng khi user nói 'security audit', 'pentest', 'kiểm tra bảo mật', 'dependency scan', hoặc trước production deploy. Deep security check /100. v2 audit format: mỗi control bind với evidence block (Command run + Output + Verdict + Evidence artifact ID).

Dùng khi user nói 'plan wave', 'cluster gaps', 'wave-pack', 'next wave', hoặc khi cần group ≥3 disjoint tasks/gaps thành 1 wave-pack chạy parallel agents. Output: wave plan markdown (foundation PR-ready) + file-overlap matrix + 3-5 agent prompts. Codifies cluster-then-spawn parallel-agent methodology (group disjoint work → spawn worktree-isolated agents → ~5x speedup vs serial PR queue).

Dùng khi user runs `/check-pr` hoặc nói 'check PR', 'verify CI', 'kiểm tra PR', 'ready to merge?'. Monitor CI via project scripts or GitHub MCP, run local tests if available, output verdict (✅ ready / ❌ fix needed / ⚠️ conditional).

Dùng khi user nói 'end session', 'đóng session', 'kết thúc session', '/end-session', 'finish work', 'wrap up', hoặc trước khi /clear cho session quan trọng. Working-tree clean + sync gate + docs-sync (gap-status / ROADMAP / wave-history / MEMORY / session-handoff) + auto-write handoff note + archive session-lock + 1-line summary. Symmetric counterpart cho /start-session.

Dùng khi user runs `/fix-pr <PR-number>` sau khi `/check-pr` flagged issues. Determines fix strategy (Case A: PR open → amend on branch; Case B: merged → follow-up PR), then addresses brainstorm/test/doc/methodology gaps.

Use when the user says 'status', 'repo status', 'tình trạng repo', 'health check', 'is the repo OK?', 'security', 'CVE', 'vuln', or when starting a conversation that needs a quick remote-repo assessment. Checks: CI, PRs/branches, audit gaps, GitHub Security (Dependabot + code-scanning + secret-scanning) → output level GREEN/YELLOW/ORANGE/RED/BLACK.

Dùng khi user runs `/start-pr` (check current PR artifacts) hoặc `/start-pr PR-X.Y feature-name` (start new PR). Verifies methodology artifacts (brainstorm, task breakdown, TDD compliance, tests, docs) and either reports status or kicks off a new feature with the methodology scaffolding.

Use when the user says 'start new session', 'bắt đầu session mới', '/start-session', 'what's the state?', 'tình trạng hiện tại', or when context must be reloaded after /clear. Loads: branch, open PRs, failing CI, recent commits, plus optional gap/wave/cloud sections when those artifacts exist. Output: single summary block. Also checks session-lock conflicts when running parallel agents.