一键导入
forge
Use when given a GitHub issue number to implement — branching, coding, verifying with architecture checks, committing with trailers, and opening a PR.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use when given a GitHub issue number to implement — branching, coding, verifying with architecture checks, committing with trailers, and opening a PR.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Write an Architecture Decision Record (MADR format) for a technical decision. Output: docs/adr/ADR-{NNN}-{slug}.md Triggers: "architecture decision", "ADR", "should we use X or Y", "technical decision", "why did we choose", "document this decision"
Design REST API contract for a feature. Output: docs/features/{slug}/api-spec.md Triggers: "design API", "REST contract", "API for X", "endpoint for X", "API contract", "write API spec", "API design"
Design database schema and Flyway migration for a feature. Output: docs/features/{slug}/db-schema.md Triggers: "design table", "db schema", "migration for X", "ERD", "flyway migration", "database schema", "schema for X", "create table"
Generate release notes and deploy checklist for a version. Output: docs/releases/v{X.Y.Z}-release.md Triggers: "release note", "release notes", "changelog", "what shipped", "write release", "deploy checklist", "prepare release", "release v"
Write a Formal SRS (IEEE 830) for a feature. Output: docs/features/{slug}/SRS.md Triggers: "write SRS", "spec feature X", "requirements for X", "write requirements", "SRS for X"
Write a comprehensive test plan for a feature: unit, integration, tenant isolation, e2e. Output: docs/features/{slug}/test-plan.md Triggers: "test plan", "test strategy", "test cases for X", "what to test", "write test plan", "test coverage for X", "how to test X"
| name | forge |
| description | Use when given a GitHub issue number to implement — branching, coding, verifying with architecture checks, committing with trailers, and opening a PR. |
Issue → PR workflow for this project. Prevents skipping architecture checks, verification evidence, commit trailers.
Invoke: /forge #N
gh issue view #N
Read: CLAUDE.md, .claude/memory/architecture.md, .claude/memory/multi-tenancy.md
Judge mode, then announce:
| Signal | Mode |
|---|---|
| Config-only, 1–2 files, no business logic | QUICK |
| New entity / service / repository layer | FULL |
| Security, auth, multi-tenancy involved | FULL |
| Design tradeoffs to document | FULL |
"Choosing FULL/QUICK because [reason]. Proceed?"
git checkout -b feature/issue-{N}-{slug}
[FULL only — CHECKPOINT]Dispatch oh-my-claudecode:planner (model: opus).
Provide: issue text + relevant code context.
Present to user:
Stop. Wait for confirm before coding.
FULL: 3-agent pipeline per task:
Executor (fresh, oh-my-claudecode:executor)
└─ implement + write tests + self-review
└─ status: DONE | DONE_WITH_CONCERNS | NEEDS_CONTEXT | BLOCKED
Spec Compliance Review (fresh, oh-my-claudecode:code-reviewer, opus)
└─ Does code match issue spec? Anything extra?
└─ FAIL → executor fixes → re-review [must pass before next]
Code Quality Review (fresh, oh-my-claudecode:code-reviewer, opus)
└─ Java 21 idioms, architecture, naming
└─ FAIL → executor fixes → re-review
Load rules/coding-style.md as executor context.
Load references/slice-anatomy.md when creating new package.
Load rules/exceptions.md when adding new exceptions.
Load references/testing-patterns.md when writing tests.
Load references/flyway-guide.md when task requires new DB columns/tables.
Load references/security-patterns.md when task touches SecurityConfig or endpoint auth.
QUICK: Implement inline. Same verification below.
Iron Law: Run command → read output → then claim. No "should pass".
Loop until all 6 pass (max 3 iterations, then escalate with evidence):
Runtime (run fresh, read actual output):
mvn test # must exit 0, 0 failures
mvn verify # includes *IT classes
Coverage (if jacoco configured):
mvn jacoco:report # service layer ≥ 60%
Static — load rules/architecture.md, run each check with Grep/AST/LSP:
Fix failures: re-dispatch fresh executor sub-agent. Never inline fix.
Load assets/commit-template.txt. Fill all trailers. Do not leave any blank.
Load assets/pr-template.md. Fill Summary + Test plan with actual numbers from Step 5.
gh pr create --title "..." --body "$(cat <<'EOF'
[filled template]
EOF
)"
mvn test this turn