omc-plan（OMC 计划）

<Use_When>

• User wants to plan before implementing -- "plan this", "plan the", "let's plan"
• User wants structured requirements gathering for a vague idea
• User wants an existing plan reviewed -- "review this plan", --review
• User wants multi-perspective consensus on a plan -- --consensus, "ralplan"
• Task is broad or vague and needs scoping before any code is written </Use_When>

<Do_Not_Use_When>

• User wants autonomous end-to-end execution -- use autopilot instead
• User wants to start coding immediately with a clear task -- use ralph or delegate to executor
• User asks a simple question that can be answered directly -- just answer it
• Task is a single focused fix with obvious scope -- skip planning, just do it </Do_Not_Use_When>

<Why_This_Exists> Jumping into code without understanding requirements leads to rework, scope creep, and missed edge cases. Plan provides structured requirements gathering, expert analysis, and quality-gated plans so that execution starts from a solid foundation. The consensus mode adds multi-perspective validation for high-stakes projects. </Why_This_Exists>

<Execution_Policy>

• Auto-detect interview vs direct mode based on request specificity
• Ask one question at a time during interviews -- never batch multiple questions
• Gather codebase facts via explore agent before asking the user about them
• Plans must meet quality standards: 80%+ claims cite file/line, 90%+ criteria are testable
• Consensus mode runs fully automated by default; add --interactive to enable user prompts at draft review and final approval steps
• Consensus mode uses RALPLAN-DR short mode by default; switch to deliberate mode with --deliberate or when the request explicitly signals high risk (auth/security, data migration, destructive/irreversible changes, production incident, compliance/PII, public API breakage) </Execution_Policy>

Mode Selection

Interview Mode (broad/vague requests)

1. Classify the request: Broad (vague verbs, no specific files, touches 3+ areas) triggers interview mode
2. Ask one focused question using AskUserQuestion for preferences, scope, and constraints
3. Gather codebase facts first: Before asking "what patterns does your code use?", spawn an explore agent to find out, then ask informed follow-up questions
4. Build on answers: Each question builds on the previous answer
5. Consult Analyst (Opus) for hidden requirements, edge cases, and risks
6. Create plan when the user signals readiness: "create the plan", "I'm ready", "make it a work plan"

Direct Mode (detailed requests)

1. Quick Analysis: Optional brief Analyst consultation
2. Create plan: Generate comprehensive work plan immediately
3. Review (optional): Critic review if requested

Consensus Mode (--consensus / "ralplan")

RALPLAN-DR modes: Short (default, bounded structure) and Deliberate (for --deliberate or explicit high-risk requests). Both modes keep the same Planner -> Architect -> Critic sequence and the same AskUserQuestion gates.

Provider overrides (supported when the provider CLI is installed):

• --architect codex — replace the Claude Architect pass with omc ask codex --agent-prompt architect "..." for implementation-heavy architecture review
• --critic codex — replace the Claude Critic pass with omc ask codex --agent-prompt critic "..." for an external review pass before execution
• If the requested provider is unavailable, briefly note that and continue with the default Claude Architect/Critic step for that stage

State lifecycle: The persistent-mode stop hook uses ralplan-state.json to enforce continuation during the consensus loop. The skill MUST manage this state:

• On entry: Call state_write(mode="ralplan", active=true, session_id=<current_session_id>) before step 1
• On handoff to execution (approval → ralph/team): Call state_write(mode="ralplan", active=false, session_id=<current_session_id>). Do NOT use state_clear here — state_clear writes a 30-second cancel signal that disables stop-hook enforcement for ALL modes, leaving the newly launched execution mode unprotected.
• On true terminal exit (rejection, non-interactive plan output, error/abort): Call state_clear(mode="ralplan", session_id=<current_session_id>) — no execution mode follows, so the cancel signal window is harmless.
• Do NOT clear during intermediate steps like Critic approval or max-iteration presentation, as the user may still select "Request changes".

Without cleanup, the stop hook blocks all subsequent stops with [RALPLAN - CONSENSUS PLANNING] reinforcement messages even after the consensus workflow has finished. Always pass session_id to avoid clearing other concurrent sessions' state.

1. Planner creates initial plan and a compact RALPLAN-DR summary before any Architect review. The summary MUST include:

- Principles (3-5) - Decision Drivers (top 3) - Viable Options (>=2) with bounded pros/cons for each option - If only one viable option remains, an explicit invalidation rationale for the alternatives that were rejected - In deliberate mode: a pre-mortem (3 failure scenarios) and an expanded test plan covering unit / integration / e2e / observability

1. User feedback *(--interactive only)*: If running with --interactive, MUST use AskUserQuestion to present the draft plan plus the RALPLAN-DR Principles / Decision Drivers / Options summary for early direction alignment with these options:

- Proceed to review — send to Architect and Critic for evaluation - Request changes — return to step 1 with user feedback incorporated - Skip review — go directly to final approval (step 7) If NOT running with --interactive, automatically proceed to review (step 3).

1. Architect reviews for architectural soundness using Task(subagent_type="oh-my-claudecode:architect",...). Architect review MUST include: strongest steelman counterargument (antithesis) against the favored option, at least one meaningful tradeoff tension, and (when possible) a synthesis path. In deliberate mode, Architect should explicitly flag principle violations. Wait for this step to complete before proceeding to step 4. Do NOT run steps 3 and 4 in parallel.
2. Critic evaluates against quality criteria using Task(subagent_type="oh-my-claudecode:critic",...). Critic MUST verify principle-option consistency, fair alternative exploration, risk mitigation clarity, testable acceptance criteria, and concrete verification steps. Critic MUST explicitly reject shallow alternatives, driver contradictions, vague risks, or weak verification. In deliberate mode, Critic MUST reject missing/weak pre-mortem or missing/weak expanded test plan. Run only after step 3 is complete.
3. Re-review loop (max 5 iterations): If Critic rejects, execute this closed loop: a. Collect all rejection feedback from Architect + Critic b. Pass feedback to Planner to produce a revised plan c. Return to Step 3 — Architect reviews the revised plan d. Return to Step 4 — Critic evaluates the revised plan e. Repeat until Critic approves OR max 5 iterations reached f. If max iterations reached without approval, present the best version to user via AskUserQuestion with note that expert consensus was not reached
4. Apply improvements: When reviewers approve with improvement suggestions, merge all accepted improvements into the plan file before proceeding. Final consensus output MUST include an ADR section with: Decision, Drivers, Alternatives considered, Why chosen, Consequences, Follow-ups. Specifically: a. Collect all improvement suggestions from Architect and Critic responses b. Deduplicate and categorize the suggestions c. Update the plan file in .omc/plans/ with the accepted improvements (add missing details, refine steps, strengthen acceptance criteria, ADR updates, etc.) d. Note which improvements were applied in a brief changelog section at the end of the plan
5. On Critic approval (with improvements applied): *(--interactive only)* If running with --interactive, use AskUserQuestion to present the plan with these options:

- Approve and implement via team (Recommended) — proceed to implementation via coordinated parallel team agents (/team). Team is the canonical orchestration surface since v4.1.7. - Approve and execute via ralph — proceed to implementation via ralph+ultrawork (sequential execution with verification) - Clear context and implement — compact the context window first (recommended when context is large after planning), then start fresh implementation via ralph with the saved plan file - Request changes — return to step 1 with user feedback - Reject — discard the plan entirely If NOT running with --interactive, output the final approved plan, call state_clear(mode="ralplan", session_id=<current_session_id>), and stop. Do NOT auto-execute.

1. *(--interactive only)* User chooses via the structured AskUserQuestion UI (never ask for approval in plain text). If user selects Reject, call state_clear(mode="ralplan", session_id=<current_session_id>) and stop.
2. On user approval (--interactive only): Call state_write(mode="ralplan", active=false, session_id=<current_session_id>) before invoking the execution skill (ralph/team), so the stop hook does not interfere with the execution mode's own enforcement. Do NOT use state_clear here — it writes a cancel signal that disables enforcement for the newly launched mode.

- Approve and implement via team: MUST invoke Skill("oh-my-claudecode:team") with the approved plan path from .omc/plans/ as context. Do NOT implement directly. The team skill coordinates parallel agents across the staged pipeline for faster execution on large tasks. This is the recommended default execution path. - Approve and execute via ralph: MUST invoke Skill("oh-my-claudecode:ralph") with the approved plan path from .omc/plans/ as context. Do NOT implement directly. Do NOT edit source code files in the planning agent. The ralph skill handles execution via ultrawork parallel agents. - Clear context and implement: First invoke Skill("compact") to compress the context window (reduces token usage accumulated during planning), then invoke Skill("oh-my-claudecode:ralph") with the approved plan path from .omc/plans/. This path is recommended when the context window is 50%+ full after the planning session.

Review Mode (--review)

1. Read plan file from .omc/plans/
2. Evaluate via Critic using Task(subagent_type="oh-my-claudecode:critic",...)
3. Return verdict: APPROVED, REVISE (with specific feedback), or REJECT (replanning required)

Plan Output Format

Every plan includes:

• Requirements Summary
• Acceptance Criteria (testable)
• Implementation Steps (with file references)
• Risks and Mitigations
• Verification Steps
• For consensus/ralplan: RALPLAN-DR summary (Principles, Decision Drivers, Options)
• For consensus/ralplan final output: ADR (Decision, Drivers, Alternatives considered, Why chosen, Consequences, Follow-ups)
• For deliberate consensus mode: Pre-mortem (3 scenarios) and Expanded Test Plan (unit/integration/e2e/observability)

Plans are saved to .omc/plans/. Drafts go to .omc/drafts/.

<Tool_Usage>

• Use AskUserQuestion for preference questions (scope, priority, timeline, risk tolerance) -- provides clickable UI
• Use plain text for questions needing specific values (port numbers, names, follow-up clarifications)
• Use explore agent (Haiku, 30s timeout) to gather codebase facts before asking the user
• Use Task(subagent_type="oh-my-claudecode:planner",...) for planning validation on large-scope plans
• Use Task(subagent_type="oh-my-claudecode:analyst",...) for requirements analysis
• Use Task(subagent_type="oh-my-claudecode:critic",...) for plan review in consensus and review modes
• CRITICAL — Consensus mode agent calls MUST be sequential, never parallel. Always await the Architect Task result before issuing the Critic Task.
• In consensus mode, default to RALPLAN-DR short mode; enable deliberate mode on --deliberate or explicit high-risk signals (auth/security, migrations, destructive changes, production incidents, compliance/PII, public API breakage)
• In consensus mode with --interactive: use AskUserQuestion for the user feedback step (step 2) and the final approval step (step 7) -- never ask for approval in plain text. Without --interactive, skip both prompts and output the final plan.
• In consensus mode with --interactive, on user approval MUST invoke Skill("oh-my-claudecode:ralph") for execution (step 9) -- never implement directly in the planning agent
• When user selects "Clear context and implement" in step 7 (--interactive only): call state_write(mode="ralplan", active=false, session_id=<current_session_id>) first, then invoke Skill("compact") to compress the accumulated planning context, then immediately invoke Skill("oh-my-claudecode:ralph") with the plan path -- the compact step is critical to free up context before the implementation loop begins
• CRITICAL — Consensus mode state lifecycle: Always deactivate ralplan state before stopping or handing off to execution. Use state_write(active=false) for handoff paths (approval → ralph/team) and state_clear for true terminal exits (rejection, error). Never use state_clear before launching an execution mode — its cancel signal disables stop-hook enforcement for 30 seconds. </Tool_Usage>

<Escalation_And_Stop_Conditions>

• Stop interviewing when requirements are clear enough to plan -- do not over-interview
• In consensus mode, stop after 5 Planner/Architect/Critic iterations and present the best version. Do NOT clear ralplan state here — the user may still select "Request changes" in the subsequent step. State is cleared only on the user's final choice (approval/rejection) or when outputting the plan in non-interactive mode.
• Consensus mode without --interactive outputs the final plan and stops; with --interactive, requires explicit user approval before any implementation begins. Always call state_clear(mode="ralplan", session_id=<current_session_id>) before stopping.
• If the user says "just do it" or "skip planning", call state_write(mode="ralplan", active=false, session_id=<current_session_id>) then MUST invoke Skill("oh-my-claudecode:ralph") to transition to execution mode. Do NOT implement directly in the planning agent.
• Escalate to the user when there are irreconcilable trade-offs that require a business decision </Escalation_And_Stop_Conditions>

<Final_Checklist>

• Plan has testable acceptance criteria (90%+ concrete)
• Plan references specific files/lines where applicable (80%+ claims)
• All risks have mitigations identified
• No vague terms without metrics ("fast" -> "p99 < 200ms")
• Plan saved to .omc/plans/
• In consensus mode: RALPLAN-DR summary includes 3-5 principles, top 3 drivers, and >=2 viable options (or explicit invalidation rationale)
• In consensus mode final output: ADR section included (Decision / Drivers / Alternatives considered / Why chosen / Consequences / Follow-ups)
• In deliberate consensus mode: pre-mortem (3 scenarios) + expanded test plan (unit/integration/e2e/observability) included
• In consensus mode with --interactive: user explicitly approved before any execution; without --interactive: plan output only, no auto-execution
• In consensus mode: ralplan state deactivated on every exit path — state_write(active=false) for handoff to execution, state_clear for terminal exits (rejection, error, non-interactive stop) </Final_Checklist>

When presenting design choices during interviews, chunk them:

1. Overview (2-3 sentences)
2. Option A with trade-offs
3. [Wait for user reaction]
4. Option B with trade-offs
5. [Wait for user reaction]
6. Recommendation (only after options discussed)

Format for each option:

### Option A: [Name]
**Approach:** [1 sentence]
**Pros:** [bullets]
**Cons:** [bullets]

What's your reaction to this approach?

Question Classification

Before asking any interview question, classify it:

Review Quality Criteria

Deprecation Notice

The separate /planner, /ralplan, and /review skills have been merged into /plan. All workflows (interview, direct, consensus, review) are available through /plan.