Notre avis
Exécute une validation pré-implantation via des agents de revue pour s'assurer que le plan est approuvé avant de coder.
Points forts
- Empêche un travail inutile en validant le plan en amont
- Processus structuré avec notation et itérations
- Détecte les lacunes dans la couverture de test, l'architecture et la complétude
Limites
- Nécessite plusieurs appels API (Oracle + agents de revue), ce qui peut être lent
- Dépend de la qualité des agents de revue (Haiku, Sonnet)
- Peut être excessif pour des changements mineurs
Lorsque vous travaillez sur des fonctionnalités complexes ou des changements nécessitant une planification minutieuse avant implémentation.
Pour des corrections de bugs triviales ou des améliorations mineures où le coût de planification n'est pas justifié.
Analyse de sécurité
SûrThe skill instructs the AI to create implementation plans and review them via sub-agents. It involves only reading and writing files to designated directories and launching review agents. There are no destructive commands, data exfiltration, or unsafe operations.
Aucun point d'attention détecté
Exemples
/countdown 42I need to validate my implementation plan for the current feature branch. Run countdown./countdown 42 (after fixing issues from previous round)name: countdown description: "Countdown - pre-implementation validation. Runs oracle (Opus) for planning, then Go Poll with review agents (Haiku) to validate before coding starts."
Countdown
The pre-launch sequence for validating a plan before implementation begins.
Usage
/countdown [issue-number]
What This Does
Countdown is the pre-implementation workflow:
- Plan - Oracle creates/refines implementation plan
- Go Poll - Review agents validate the plan
- Iterate - Fix issues until all agents report GO
- Launch - Begin implementation only when validated
Sequence
Phase 1: Oracle Planning
Oracle creates the implementation plan at:
~/.claude/plans/reovim/{issue_number}-{subject}.md
Phase 2: Go Poll
Review agents validate the plan:
| Agent | Model | Focus | |-------|-------|-------| | mission-control | Haiku | Plan completeness, phases, acceptance criteria | | telemetry | Haiku | Test strategy, coverage targets | | flight-director | Sonnet | Architecture, Unix philosophy, layer boundaries |
A+ Skip Rule
If an agent gives ALL A+ in round N, skip that agent in round N+1.
Output Structure
tmp/{ISSUE}/
└── countdown/
└── round-{N}/
├── mission-control.md
├── telemetry.md
└── flight-director.md
Grading Scale
| Grade | Meaning | |-------|---------| | A+ | Exemplary plan (skip next round) | | A | Ready for implementation | | B | Minor gaps - refine plan | | C | Significant gaps - more planning needed | | F | Major issues - rethink approach |
Success Criteria
Must achieve A+ from ALL THREE review agents before implementation.
If any agent gives below A+:
- Refine the plan based on feedback
- Re-run
/countdown - Repeat until all agents report "go"
Instructions
When invoked, you MUST:
- Determine the issue number from context (git branch, user input, or ask)
- Check if a plan exists at
~/.claude/plans/reovim/{issue}-*.md - If no plan exists or plan needs refinement:
- Launch
oracleagent (Opus) to create/refine the plan
- Launch
- Determine the round number (start at 1, increment for re-reviews)
- Check previous round grades - skip agents with A+ from prior rounds
- Create the output directory:
tmp/{ISSUE}/countdown/round-{N}/ - Launch review agents IN PARALLEL using Task tool with
model: "haiku" - After all agents complete, summarize the grades in a table
- If any grade is below A, list what needs to be fixed in the plan
Phase 1: Oracle (if needed)
Create/refine implementation plan for issue #{ISSUE}.
You are Oracle - the far-seeing architect.
1. Understand the requirements from the issue
2. Explore the codebase to understand current architecture
3. Design a phased implementation plan
Write the plan to: ~/.claude/plans/reovim/{ISSUE}-{subject}.md
Include:
- Summary and approach
- Phases with specific files and changes
- Risks and mitigations
- Test strategy
- Acceptance criteria
Phase 2: Review Agents
Agent 1 (mission-control):
Countdown review for issue #{ISSUE} (Round {N}).
You are Mission Control validating the plan before launch.
Focus: Plan completeness and clarity.
1. Read the plan at ~/.claude/plans/reovim/{ISSUE}-*.md
2. Verify all phases have clear acceptance criteria
3. Check dependencies between phases
4. Ensure scope is well-bounded
Write your report to: tmp/{ISSUE}/countdown/round-{N}/mission-control.md
End with: "Mission Control: GO / NO-GO" and grade (A+/A/B/C/F).
Agent 2 (telemetry):
Countdown review for issue #{ISSUE} (Round {N}).
You are Telemetry validating the test strategy.
Focus: Test coverage planning.
1. Read the plan at ~/.claude/plans/reovim/{ISSUE}-*.md
2. Verify test strategy covers happy paths, errors, edge cases
3. Check that critical functionality has test targets
4. Identify any testing gaps
Write your report to: tmp/{ISSUE}/countdown/round-{N}/telemetry.md
End with: "Telemetry: GO / NO-GO" and grade (A+/A/B/C/F).
Agent 3 (flight-director):
Countdown review for issue #{ISSUE} (Round {N}).
You are Flight Director validating the architecture.
Focus: Unix philosophy and layer boundaries.
1. Read the plan at ~/.claude/plans/reovim/{ISSUE}-*.md
2. Verify mechanism vs policy separation
3. Check layer boundaries are respected
4. Assess complexity and simplicity
Write your report to: tmp/{ISSUE}/countdown/round-{N}/flight-director.md
End with: "Flight Director: GO / NO-GO" and grade (A+/A/B/C/F).
CRITICAL:
- Launch
oraclewithmodel: "opus"for planning - Launch review agents with
model: "{model}"for validationmodel: "haiku"for fast, efficient reviewsmodel: "sonnet"for comprehensive, precise reviews
- Skip agents that received ALL A+ in previous rounds
- Do NOT start implementation until all agents report GO
Expert Next.js App Router
Developpement
Un skill qui transforme Claude en expert Next.js App Router.
Générateur de README
Developpement
Crée des README.md professionnels et complets pour vos projets.
Rédacteur de Documentation API
Developpement
Génère de la documentation API complète au format OpenAPI/Swagger.