Install
$ agentstack add skill-dachent-skills-deep-planning-codex ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Codex Deep Planning Orchestrator
> Deprecated for GPT-5.6 Sol. Use native Codex Plan Mode, with focused verification or adversarial review skills when needed. Do not use this package for new work.
Overview
Run a Codex-native deep planning protocol that adapts to software with Git, software without Git, business artifact work, and mixed business-coding projects. Produce durable planning artifacts, proof criteria, and a reviewed execution strategy before implementation.
Core Contract
- Treat the user's invocation text as the planning objective.
- Create or reuse the target project's
.deep-planning/workfolder. - During planning, do not edit implementation files. Allowed planning writes are
.deep-planning/,.ultraplan/plan.md, approved design/plan docs,CONTEXT.md, and accepted ADRs. - Every phase updates
.deep-planning/state.md. - Every phase ends with exactly one status:
READY_FOR_PROCEED,BLOCKED_NEEDS_USER_DECISION,BLOCKED_BY_MISSING_EVIDENCE, orFAILED_VALIDATION. - Stop at explicit proceed gates. Continue only after the user says
PROCEEDor gives equivalent explicit approval. - Use companion skills when available:
$repo-map-codex,$verification-plan-codex,$adversarial-plan-review-codex,$grill-with-docs-codex,$grill-me-codex,$handoff-codex,$ultraplan-codex,superpowers:brainstorming,superpowers:systematic-debugging,superpowers:writing-plans, andsuperpowers:verification-before-completion. If unavailable, perform the same behavior locally. - Use Codex subagents only when explicitly allowed by the user or current session instructions. Otherwise perform bounded local exploration and review.
Companion Skill Contract
When a companion skill is invoked inside this workflow, require a Deep Planning Delta and merge it into .deep-planning/state.md before any proceed gate:
## Deep Planning Delta
Phase: ...
Status: READY_FOR_PROCEED | BLOCKED_NEEDS_USER_DECISION | BLOCKED_BY_MISSING_EVIDENCE | FAILED_VALIDATION
Updated artifacts:
- ...
Blockers:
- ...
Open assumptions:
- ...
Next action:
...
If the companion skill writes a native artifact outside .deep-planning/, such as .ultraplan/plan.md, keep that native artifact and also update or link the corresponding .deep-planning/ artifact so the validator can reason over the full planning packet.
References
Read references only when needed:
references/project-modes.mdwhen selecting or revising project mode.references/artifact-contracts.mdbefore writing or validating.deep-planning/artifacts.references/phase-contracts.mdbefore running the full workflow or resuming a workflow.references/proceed-gates.mdbefore stopping for approval.references/subagent-strategy.mdbefore delegating planning or execution work.
Phase Flow
- Harness preflight: record workspace, writable roots, Git state if present, available skills, subagent authorization, allowed artifact paths, and selected project mode.
- Intake and criteria: sharpen success criteria, failure criteria, out-of-scope boundaries, owner decisions, risk tolerance, and documentation permissions.
- Project map and evidence catalog: invoke
$repo-map-codexif available or produce equivalentrepo-map.mdandevidence-catalog.md. - Failure autopsy: if prior failures exist, use systematic debugging or diagnosis to build
failure-autopsy.mdanddead-ends-registry.md. - Assumption attack: challenge criteria, evidence, dead ends, and unresolved claims; update
assumption-ledger.md. - Targeted probes: run low-cost probes mapped to assumptions. Do not implement features.
- Design synthesis: use brainstorming only if real design ambiguity remains; otherwise write the selected design directly.
- Grounded implementation plan: use
$ultraplan-codexfor software-heavy work; for artifact-heavy work write.deep-planning/implementation-plan.md. When$ultraplan-codexwrites.ultraplan/plan.md, also write or update.deep-planning/implementation-plan.mdwith the execution sequence, verification command, and a pointer to the native plan. - Verification plan: invoke
$verification-plan-codexif available or produce equivalent proof criteria. - Adversarial review: invoke
$adversarial-plan-review-codexif available or locally red-team the plan and fix blockers. - Final execution plan: use
superpowers:writing-plansfor code-heavy work; otherwise write a decision-complete artifact execution plan. - Execution strategy: choose subagent-driven, parallel, or inline execution only after explicit approval.
- Final verification and archive: use verification-before-completion behavior, then write final validation and handoff records.
Project Modes
Classify the project in preflight:
| Mode | Use when | Delivery primitive | | --- | --- | --- | | software-git | Git repository and commits/PRs are allowed | branch, commit, PR, CI | | software-no-git | Code exists but no commit workflow is allowed | file snapshots, milestone state, validation log | | business-artifact | Work is mostly docs, reports, spreadsheets, decisions, or analysis | approval packet, evidence log, deliverable folder | | mixed-business-coding | Business deliverables include code, scripts, data, or automation | artifact snapshots plus code and business validation |
Default to mixed-business-coding when uncertain.
Required Artifacts
Maintain these in the target project's .deep-planning/ folder as applicable:
state.mdharness-preflight.mdproject-mode.mdsuccess-criteria.mdfailure-criteria.mdout-of-scope.mdstakeholder-map.mdrepo-map.mdevidence-catalog.mdassumption-ledger.mddecision-log.mdfailure-autopsy.mddead-ends-registry.mdprobe-results.mddesign.mdimplementation-plan.mdverification-plan.mdadversarial-review.mdexecution-strategy.mdfinal-validation.mdhandoffs/
Use .ultraplan/plan.md when $ultraplan-codex is part of the selected software-heavy workflow, but do not treat it as a replacement for .deep-planning/implementation-plan.md.
Handoff Policy
Do not run full handoff after every small phase by default. Always update .deep-planning/state.md. Run $handoff-codex at major context gates: after Phase 0, after failure autopsy, after final plan approval, before execution handoff, and at final archive. In this workflow, store handoffs under .deep-planning/handoffs/.md or record a state pointer to the handoff path.
Validation
Before presenting a final execution plan, run scripts/validate_deep_planning_artifacts.py against the target project when possible. Treat missing required artifacts, incomplete state fields, unresolved proceed gates, failed adversarial verdicts, blocking review findings, or unreviewed verification criteria as blockers.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: dachent
- Source: dachent/skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.