Menu
Research-validated planning - research -> plan -> verify cycle for high-confidence implementation plans
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
6-stage structured development cycle with stage-based tool restrictions
Multi-angle release quality verification using parallel expert review teams
Full Self Driving — autonomous release loop that processes all auto-dev-eligible GitHub issues until none remain, by repeatedly running /pipeline auto-dev then /homework.
Invoke and resume YAML-defined pipelines by name — /pipeline auto-dev runs the full release pipeline
Analyze release workflow findings and recommend follow-up actions — execute immediately or register as issues
Auto-detect project context and optimize harness — deactivate unused agents/skills, suggest missing experts, generate project profile
| name | deep-plan |
| description | Research-validated planning - research -> plan -> verify cycle for high-confidence implementation plans |
| scope | core |
| context | fork |
| version | 1.1.0 |
| user-invocable | true |
| argument-hint | <topic-or-issue> |
| teams-compatible | true |
Research-validated planning that eliminates the gap between research assumptions and actual code. It runs Discovery Research, Reality-Check Planning, and Plan Verification before handing off implementation.
Full phase detail: guides/deep-plan/phases.md
/deep-plan <topic-or-issue>
/deep-plan "implement caching layer for API responses"
/deep-plan #325 new authentication system
| Phase | Name | Key Activity |
|---|---|---|
| 1 | Discovery Research | Comprehensive research with ADOPT / ADAPT / AVOID taxonomy |
| 2 | Reality-Check Planning | Explore current code and keep only real gaps |
| 3 | Plan Verification | Feasibility, duplication, and test-risk verification |
| Rule | Detail |
|---|---|
| Phase 1 | Invoke or inline the research workflow, depending on runtime context |
| Phase 2 | Use up to 3 parallel Explore agents for current-code verification |
| Phase 3 | Use up to 3 focused verification agents |
| Orchestrator owns synthesis | The main conversation manages phase transitions and verdicts |
| REVISE limit | Max 2 cycles before user escalation |
| Skill | Scope | Code Verification |
|---|---|---|
/research | Analysis only | None by default |
| Plan mode | Planning only | Yes |
/structured-dev-cycle | Implementation pipeline | Yes, stage by stage |
/deep-plan | Analysis + planning + verification | Yes, 3-pass cross-verification |
Codex-native verification artifacts go under .codex/outputs/sessions/YYYY-MM-DD/deep-plan-HHmmss.md and may be written with the repo's normal file-write APIs.
If any research, planning, or verification delegate must touch .claude/**, .claude/outputs/**, or templates/.claude/**, include this block in that delegate prompt:
Sensitive-path compatibility note:
1. Keep Codex-native .codex/** artifacts on the normal file-write path.
2. On Claude Code v2.1.121+ with bypassPermissions, direct writes to .claude/skills/, .claude/agents/, and .claude/commands/ are allowed.
3. On Claude Code v2.1.126+ with bypassPermissions, broader protected paths such as .claude/**, .git/**, and .vscode/** are also covered.
4. Use /tmp/deep-plan-<timestamp>.md only as a legacy fallback when the runtime is older or still prompts, then verify the resulting diff.
The protocol must be copied into delegate prompts. A parent-only mention is insufficient because forked contexts can lose the directive.
When Agent Teams are available and the plan has multiple independent verification lanes, the Phase 3 agents may run as a coordinated team. Otherwise, use bounded parallel subagents and aggregate results in the main conversation.
After PASS, return an implementation-ready plan with file scopes, dependencies, test commands, release gates, rollback notes, and any unresolved risks.
Claude compatibility Agent calls only (R010 “Delegated Permission Ownership”): Pass mode: "bypassPermissions" when the active Claude session uses bypass permissions. Native Codex spawn_agent has no mode parameter; use the installed agent_type and active Codex runtime permissions instead.