teamwork-plan

# Teamwork Plan

## Outcome

Turn evidence into an executable scope with falsifiable verification and stop
boundaries.

## Enter When

Use when planning is requested or non-trivial action lacks an accepted plan.
Route unresolved external behavior or architecture to research, and unknown
reproducible failures to debug. Tiny fixed-scope work may go to execution.

## Do And Boundaries

Resolve decision-critical gaps and map requirements to evidence or proof. Define
scope and protected surfaces; for code, identify owner/flow, tests/config, and
invariants before choosing the smallest direct change.

A plan is simple only when scope/risk are fixed and no material user decision
remains. Every other plan enters evidence-first `grill-me`, independent of file
count, and needs a confirmed Decision Summary. Confirmation grants no execution.

Source critical values from the user, repository, or justified derivation.
Leave unresolved values open and block only dependent work; recommendations are
not facts.

Write owned actions, dependencies, verification, and stop/replan conditions
sufficient to execute without redesign. Delegate only independent tracks. Use a
table or diagram only when it materially clarifies comparison or flow. Use a
durable plan for cross-turn, high-risk, delegated, public-contract, or goal work.

## Done When

The plan states goal, evidence, scope, steps, proof, risks, stop conditions, and
next action. Each requirement maps to an action and proof; each phase names its
surface, inputs, outputs, dependencies, and replan trigger. Acceptance requires
the user or governing workflow.

## Escalate

Ask when a user decision changes behavior, architecture, risk, or acceptance.
Return to research/debug when evidence cannot support a plan.

## Conditional Protocols

In Codex Plan mode, read `skills/using-teamwork/references/plan-output.md` and
apply its native bridge and readiness gate, including the decision gate, before the
authoritative plan. Use `verification-patterns.md` for proof,
`subagent-dispatch.md` and `subagent-contract.md` for delegation, and
`goal-iteration.md` for retry invariants. Paths are under
`skills/using-teamwork/references/`.