dp-campaign-control

name	dp-campaign-control
description	Operate SPEC-80-style campaign work in dp-codex-aware repositories; trigger on campaign, next goal, recover, block, verify, handoff.

DP Campaign Control

Respect AGENTS.md and any nested AGENTS.md before using this skill.

Use dp as the campaign control plane. Codex may reason and edit, but dp owns state, evidence, blocking gates, and recovery artifacts.

Start

1. Confirm the repo opts into dp with .beads/, dp-policy.json, or repo instructions.
2. Run dp doctor --json.
3. If a Beads issue is the intake surface, run dp task claim --json or claim the known issue.
4. If a campaign is the intake surface, run dp campaign recover docs/campaigns/<campaign>.json --json.
5. For the next campaign goal, prefer dp campaign run docs/campaigns/<campaign>.json --driver codex --supervised --managed --json.

The managed campaign run claims at most one goal, emits the handoff package, and stops. It does not spawn Codex, execute evidence, or mark work complete.

Work a Goal

1. Read the handoff's read_first files before editing.
2. Start the goal with dp goal start docs/goals/<goal>.json --agent codex --json.
3. Stay inside the handoff boundaries and GoalContract allowed paths.
4. Run the smallest relevant check first.
5. On failure, use the recovery classification from dp-failure-recovery. A failed gate blocks completion, not diagnosis or authorized repair.
6. Repair local failures before broadening to make check; do not repeat unchanged attempts or weaken evidence.
7. Use dp evidence run docs/evidence/<evidence>.json --output docs/evidence-runs/<run>.json --json when the goal has an evidence plan.
8. Use dp verify --goal docs/goals/<goal>.json --evidence docs/evidence-runs/<run>.json --json before claiming completion.
9. Use dp campaign sync-beads docs/campaigns/<campaign>.json --write --json when campaign state needs to reconcile back to Beads.

Block or Release

Block through dp when a required spec, decision, validator, or scope change is missing:

dp goal block docs/goals/<goal>.json --reason needs_decision --write-artifact --json

Use the GoalContract blocker routes: needs_specification, needs_decision, needs_validator, unsafe_scope, or budget_exhausted.

Release the goal instead of leaving an invisible claim when stopping without completion:

dp goal release docs/goals/<goal>.json --reason "context reset" --json

Close Out

Before ending the session:

1. Run the focused tests for the changed slice.
2. Run make check.
3. Run dp doctor --json.
4. Run bd --readonly status --json.
5. Close or update the Beads issue with evidence.
6. Commit and push.
7. Confirm git status reports the branch is up to date with origin.

Boundaries

Do not mark completion from agent narration. Do not execute raw shell from generated JSON. Do not call an LLM from hooks, validators, evidence assertions, or CI. Do not spawn Codex from dp in this workflow. Do not replace Beads with campaign files; synchronize through explicit dp or bd commands.