Execute → Verify → Fail? → Fix → Repeat (up to max iterations)
1---2name: oh-my-ccg-impl3description: RPI 实施阶段 — 按零决策计划机械执行4---56# oh-my-ccg Implementation Phase
78You are executing the RPI Impl phase. Execute the plan mechanically.
910## Prerequisites
11- RPI state must be in PLAN phase (or resuming IMPL)
12- Tasks must be defined in state/artifacts
1314## Workflow
1516### Step 1: Restore State
17Use the **oh-my-ccg-tools** MCP server's `rpi_state_read` tool.
18Load the task list and current progress.
1920### Step 2: Assess Parallelism
21Check task dependencies:
22- If multiple independent tasks exist → use Team mode for parallel execution
23- If tasks are sequential → execute one by one with executor agent
2425### Step 3: Execute Tasks
26For each task (or batch of parallel tasks):
27281. **Route to appropriate model for prototype** (via MCP tools):
29 - Frontend tasks → call `ask_gemini(agent_role="designer")` for prototype, then executor rewrites to production
30 - Backend tasks → call `ask_codex(agent_role="architect")` for prototype, then executor rewrites to production
31 - General tasks → executor implements directly (no external model needed)
32332. **Executor implements**:
34 - Read task spec completely
35 - Explore existing code patterns
36 - Rewrite prototype to production quality (NEVER apply external model output directly)
37 - Run type checks and tests
38393. **Side-effect review** (MANDATORY before moving to next task):
40 - Verify changes stay within task scope
41 - Check no unintended dependencies affected
42 - Confirm interfaces match expectations
4344### Step 4: Cross-Review
45After all tasks complete, run dual-model cross-review using MCP tools **in parallel**:
4647- `ask_codex(agent_role="code-reviewer", background=true)` — reviews logic/security
48- `ask_gemini(agent_role="designer", background=true)` — reviews patterns/maintainability
4950Collect both results via `check_job_status`, then:
51- Fix any Critical findings immediately
52- Log Warnings for review phase
5354### Step 5: Verify
55Run verifier agent:
56- All tests pass
57- Build succeeds
58- No type errors
59- Constraints satisfied
6061### Step 6: Update State
62Use `rpi_state_write` MCP tool to:
63- Mark tasks complete in state
64- Transition phase to "impl"
65- If all tasks done → prompt for `/oh-my-ccg:review`
6667### Ralph Integration
68If ralph mode is active:
69- Execute → Verify → Pass? → Complete
70- Execute → Verify → Fail? → Fix → Repeat (up to max iterations)
Run npx skillmds add majiayu000/oh-my-ccg-impl in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
RPI 实施阶段 — 按零决策计划机械执行 It is listed under Coding & Dev Tools on SkillMD.
This skill has not completed SkillMD's automated safety review yet. Capability flags: docs only. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
majiayu000 (@majiayu000) published this skill. Their other Agent Skills are listed on their SkillMD profile.