Menu
Run plan.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Dispatch fresh-context refuters (model-agnostic; multi-model opt-in) to attack a completion claim at the shared-trunk pawl before landing. Triggers: pre-land validation, refute.
Dispatch fresh-context refuters (model-agnostic; multi-model opt-in) to attack a completion claim at the shared-trunk pawl before landing. Triggers: pre-land validation, refute.
Run pre mortem.
Stress-test plans before work. Use when: a plan is drafted but not yet executed and you want to surface failure modes, risks, and what would prove it wrong before committing.
Repair skill hygiene. Triggers: "heal-skill", "heal skill", "repair skill hygiene.".
Repair skill hygiene. Triggers: "heal-skill", "heal skill", "repair skill hygiene.".
| name | plan |
| description | Run plan. |
Quick ref: turn a goal or research artifact into
.agents/plans/*.md, optional br issues, dependency waves, file ownership, and validation checks.
Execute this workflow. Do not only describe it. Keep planning separate from
implementation. A finished plan should let $crank, $implement, or a future
Codex session execute without chat-only context.
Given $plan <goal> [--auto]:
| Flag | Purpose |
|---|---|
--auto | Skip the human approval gate for $rpi and other autonomous chains |
--fast-path | Force the minimal 1-2 issue plan shape |
--deep | Force symbol-level/deep plan detail |
--skip-symbol-check | Skip symbol verification for greenfield plans |
--skip-audit-gate | Skip baseline audit gate for docs-only plans |
If br is unavailable, still write the markdown plan in .agents/plans/.
Use the Skill Ports and Adapters vocabulary for the boundary from Discovery into Plan:
| Boundary piece | Plan contract |
|---|---|
| Inbound port | plan_slices from BDD intent, bead, research artifact, or execution packet |
| Outbound ports | persist_issue, verify_symbols, retrieve_context, seed_execution_packet |
| Driving adapter | $plan skill invocation |
| Driven adapters | br, bv, rg, .agents/findings, .agents/plans, execution-packet writer |
| Context packet | slice plan, file dependency matrix, acceptance criteria, test levels |
| Guard adapter | stale-scope verification, symbol verification, wave-validity check |
Feature: Plan converts dense intent into executable slices
Scenario: Plan consumes Discovery output
Given Discovery provides density fields and artifact links
When Plan receives the `plan_slices` port request
Then each slice has acceptance criteria, write scope, test levels, and ownership
And no slice depends on raw Discovery chat context
ao beads verify <bead-id> before decomposition. Do not plan against stale
citations without revalidation..agents/plans/ and locate prior research,
handoffs, findings, planning rules, and relevant .agents/ history..agents/planning-rules/*.md; fall back
to .agents/findings/registry.jsonl. Treat active findings as hard
planning context. Record applied finding IDs in the plan with an
Applied findings: line, even when the value is none.$council --mode=debate --focus=ideas) before decomposition. Keep it
advisory, not mandatory. Skip it for single-session or non-contested plans.
Evidence from the 2026-05-17 Mt Olympus run: roughly 22 min wall-clock,
3/5 operator defaults flipped, and one already-shipped adapter bug surfaced.file:line, test
locations, and package/import relationships.--deep.files_exist, content_check, command, tests, or
lint). Every bead MUST also carry an embedded ## Scenarios Gherkin
block (Given/When/Then) — by default, without being asked. Free-text-only
acceptance is invalid (AGENTS.md); promote any free text to scenarios before
creating the bead. The ## Scenarios block is the behavior layer and sits
above the acceptance_criteria YAML (the machine-checkable layer); they are
complementary, never substitutes. One scenario per distinct Given/When/Then
behavior. Non-trivial plans and bead bodies should include the hexagon:
boundary block: inbound port, bounded context, adapters, context packet, and
done state.bash scripts/refresh-codex-artifacts.sh --scope worktree
in verification..agents/plans/YYYY-MM-DD-<goal-slug>.md and the
template in references/plan-document-template.md.--auto; otherwise ask whether to
proceed, revise, or return to research.Every non-trivial plan must include:
Read references/plan-document-template.md for the canonical shape.
additionalProperties: false, put schema work
before consumers in an earlier wave.User says: $plan "add rate limiting"
Produce a plan with file inventory, issues, validation, and wave order.
User says: $plan --auto ".agents/research/auth.md"
Use the research as input, write the plan, create tracking tasks when possible,
and skip the approval gate.
Read references/examples.md for full examples.
| Problem | Response |
|---|---|
| br is missing | Write the markdown plan and note that issue creation was skipped |
| Prior research is thin | Explore enough to produce file and symbol evidence |
| Same file appears in parallel issues | Serialize or merge those issues before handoff |
| Baseline audit is missing | Mark the plan incomplete unless --skip-audit-gate is justified |