campaign

Orchestrates multi-workstream initiatives by surveying, planning, dispatching tiered workers, and reconciling landed work.

10|2|Updated Jan 24, 2026
One-click install
npx skills add https://github.com/nrdxp/predicate --skill campaign-nrdxp
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: campaign
Source: https://github.com/nrdxp/predicate/tree/main/skills/campaign
Command: npx skills add https://github.com/nrdxp/predicate --skill campaign-nrdxp

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Long-horizon, multi-workstream engineering initiatives drift when a single agent tries to do everything: premises baked into worker prompts go stale, workers self-certify their own output, and expensive model tokens get spent on routine execution. This Skill defines a closed-loop campaign protocol that keeps an architect-tier council in control of planning and judgment while cheap-tier workers execute. ## Core Features & Use Cases - Exhaustive Survey & Planning: Fans out cheap-tier survey subagents across orthogonal risk angles, records an evidence-grounded findings ledger, and derives a dependency DAG of worker boundary conditions (IBCs) validated by Nickel contracts. - Tiered Orchestration & Dispatch: Routes each DAG node to the cheapest capable model tier, authors worker IBCs just-in-time with premise freshness re-verified against the current HEAD, and enforces readiness metrics (evaluator coverage, seam completeness, red-test inventory) before dispatch. - Reconciliation & Judgment: Re-runs acceptance evaluators on every landed changeset, checks surface honesty and coherence impact, and issues ACCEPT, REWORK, QUARANTINE, or ESCALATE verdicts with tiered DAG-amendment governance. - Use Case: Running a production-readiness assessment across a large codebase: the council surveys for risks, plans a mitigation DAG, dispatches autonomous workers per node, and judges each merged result until all findings are mitigated or accepted as risk. ## Quick Start Invoke the campaign workflow with an approved boundary document by asking the agent to run /campaign on your approved IBC to survey, plan, and orchestrate the initiative.

Frequently Asked Questions about campaign

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a multi-workstream campaign workflow with an AI agent?

Start from an approved IBC (boundary document) and invoke the /campaign workflow. The council surveys the target system with cheap-tier subagents, builds a mitigation DAG, dispatches worker IBCs per node, and reconciles each landed changeset until all findings are resolved.

When should I use the campaign workflow instead of a single core workflow?

Use campaign only for multi-workstream initiatives where survey, tier routing, and reconciliation pay for themselves. A single well-scoped change should go straight to the core workflow, and single-artifact polish to refine, because campaign orchestration overhead is itself a token cost.

What is premise freshness in a campaign DAG?

Premise freshness means re-verifying each pending node's stated world-state assumptions (S1 tripwires) against the current HEAD before dispatch. A stale premise marks the node INVALIDATED and its IBC is realigned, preventing cross-node drift from accumulating.

How are campaign artifacts validated and stored?

The findings ledger, campaign DAG, context map, and reconcile log are authored as YAML under .ledger/state/ and validated with nickel export --apply-contract against locked Nickel contracts. Narrative working files live in gitignored .scratch/, while state is committed in the ledger sub-repository.

What happens when a campaign worker's changeset fails review?

Failed changesets receive a REWORK verdict with a corrective delta IBC and are re-dispatched. If an acceptance evaluator cannot run, the node is QUARANTINED until the evaluator is restored; structural plan faults trigger ESCALATE back to PLAN or ORCHESTRATE.

Can campaign DAG nodes be changed after the plan is approved?

Yes, amendments are expected and classified by contract. Mechanical amendments (surface changes, realignment, splits, merges, sequencing) proceed on the architect's recorded assent, while load-bearing changes to goals or acceptance criteria require head ratification before re-dispatch.