Full-Cycle Review

---

name: full-cycle description: This skill should be used when the user asks for a "full review and fix", "find and fix issues", "review and implement", "end-to-end review", "audit and fix", "comprehensive review", "full-cycle review", or wants experts to find issues AND have them automatically implemented (not just reviewed). Use /mira:experts instead if the user only wants opinions or analysis without code changes. argument-hint: "[focus area or --discovery-only]"

Full-Cycle Review

Requires: Claude Code Agent Teams feature.

End-to-end expert review with automatic implementation and QA verification.

Arguments: $ARGUMENTS

Instructions

1. Parse arguments (optional):

   • --discovery-only -> Only run Phase 1 (same as /mira:experts)
   • --skip-qa -> Skip Phase 3 QA verification
   • --members nadia,sable -> Only spawn these specific discovery experts (by first name)
   • Any other text -> use as the context/focus for the review

2. Determine context: The user's question, the area to review, or the scope of analysis. If no context is obvious, ask the user what they'd like reviewed.

---

Phase 1: Discovery

3. Launch discovery team: Call the Mira launch MCP tool to get agent specs:

   launch(team="expert-review-team", scope=user_context, members="nadia,sable" or omit for all)
   

   The members parameter is only needed if the user passed --members.

4. Create the team:

   TeamCreate(team_name=result.data.suggested_team_id)
   

5. Create and assign discovery tasks: For each agent in result.data.agents:

   TaskCreate(subject=agent.task_subject, description=agent.task_description)
   TaskUpdate(taskId=id, owner=agent.name, status="in_progress")
   

6. Spawn discovery experts: For each agent in result.data.agents, use the Task tool:

   Task(
     subagent_type="general-purpose",
     name=agent.name,
     model=agent.model,
     team_name=result.data.suggested_team_id,
     prompt=agent.prompt + "\n\n## Context\n\n" + user_context,
     run_in_background=true
   )
   

   Spawn all discovery experts in parallel (multiple Task calls in one message).

   IMPORTANT: Do NOT use mode="bypassPermissions" for discovery agents -- they are read-only explorers. IMPORTANT: Always pass model="sonnet" to the Task tool. This ensures read-only agents use a cost-efficient model.

7. Wait for findings: All discovery experts will send findings via SendMessage. Wait for all to finish, then shut them down.

---

Phase 2: Synthesis + Implementation

8. Synthesize findings into a unified report:

   • Consensus: Points multiple experts agree on
   • Key findings per expert: Top findings from each specialist
   • Tensions: Where experts disagree -- present both sides with evidence
   • Prioritized action items: Concrete fixes grouped by file ownership

   IMPORTANT: Preserve genuine disagreements. Do NOT force consensus.

9. Present synthesis to user and WAIT for their approval before proceeding to implementation. Do not auto-proceed.

10. Launch implementation team: Call launch to get implementation agent specs:

    launch(team="implement-team", scope=approved_items)
    

11. Spawn Kai (implementation planner) from the launch results:

    Task(
      subagent_type="general-purpose",
      name="kai",
      model=kai_agent.model,
      team_name=implement_result.data.suggested_team_id,
      prompt=kai_agent.prompt + "\n\n## Approved Findings\n\n" + approved_items,
      run_in_background=true
    )
    

    Kai groups fixes by file ownership, identifies dependencies, and sets max 3-5 fixes per agent.

12. Spawn implementation agents based on Kai's work breakdown:

    Task(
      subagent_type="general-purpose",
      name="fixer-{group-name}",
      team_name=implement_result.data.suggested_team_id,  # result from step 10's launch call
      prompt=implementation_prompt + task_descriptions,
      run_in_background=true,
      mode="bypassPermissions"
    )
    

    Follow the implement-team coordination rules: strict file ownership, max 3-5 fixes per agent, schema changes first, verify with cargo test --no-run (NEVER --release). Spawn all implementation agents in parallel. Monitor build diagnostics and send hints if needed.

13. Spawn Rio (integration verifier) from the launch results after implementation agents complete:

    Task(
      subagent_type="general-purpose",
      name="rio",
      model=rio_agent.model,
      team_name=implement_result.data.suggested_team_id,
      prompt=rio_agent.prompt + "\n\n## Changes Made\n\n" + summary_of_changes,
      run_in_background=true,
      mode="bypassPermissions"
    )
    

    Rio runs compilation checks, linters, tests, and fixes cross-agent issues.

14. Wait for implementation: All agents report completion via SendMessage. Shut them down.

---

Phase 3: QA Verification

15. Launch QA team: Call launch to get QA agent specs:

    launch(team="qa-hardening-team", scope=summary_of_changes)
    

16. Spawn QA agents: For each agent in the QA launch results:

    Task(
      subagent_type="general-purpose",
      name=agent.name,
      model=agent.model,
      team_name=qa_result.data.suggested_team_id,  # result from step 15's launch call
      prompt=agent.prompt + "\n\n## Changes Made\n\n" + summary_of_changes,
      run_in_background=true
    )
    

    IMPORTANT: Do NOT use mode="bypassPermissions" for QA agents -- they are read-only. IMPORTANT: Always pass model="sonnet" to the Task tool. This ensures read-only agents use a cost-efficient model.

17. Create and assign QA tasks for each auditor.

18. Wait for QA results: If issues found, either fix directly or spawn additional fixers.

---

Phase 4: Finalize

19. Verify final build: cargo clippy --all-targets --all-features -- -D warnings + cargo fmt --all -- --check + cargo test (NEVER --release).
20. Shut down all remaining agents.
21. Report final summary to user with all changes made.
22. Cleanup: TeamDelete

Handling Stalled Agents

If an agent has not responded after an unusually long time, send it a direct message via SendMessage to check status. For discovery agents, shut down if unresponsive and note the gap. For implementation agents, fix directly or reassign. Do not wait indefinitely.

Examples

/mira:full-cycle
-> Prompts for what to review, then runs full discovery -> implementation -> QA cycle

/mira:full-cycle Review the database layer for issues
-> 4 experts review the DB layer, findings are implemented, QA verifies

/mira:full-cycle --discovery-only
-> Only runs Phase 1 (equivalent to /mira:experts)

/mira:full-cycle --skip-qa
-> Runs discovery + implementation but skips QA phase

/mira:full-cycle --members nadia,jiro
-> Only Nadia and Jiro run discovery, then full implementation + QA cycle

Phases and Agents

Phase	Agents	Purpose
Discovery	Nadia, Jiro, Sable, Lena (expert-review-team)	Find issues, propose improvements
Implementation	Kai plans, dynamic agents execute, Rio verifies (implement-team)	Implement fixes in parallel
QA	Hana, Orin, Kali, Zara (qa-hardening-team)	Verify changes, catch regressions