skills
User GuideConsensus

Plan

plan produces a structured markdown plan from a goal and optional inline constraints. It is the consensus skill for planning from no existing plan: two provider-CLI-backed peers draft plans from the goal, the wrapper can synthesize a merged plan, and the final markdown file includes steps, dependencies, risks, a deliberation log, and resolution metadata.

For an operator walkthrough of goal input, expected JSONL, sidecar output, and resolution review, see the consensus plugin README and skills/plan/references/operator-qa.md.

Plan from a goal

Use an inline goal and optional inline constraints:

node plugins/consensus/skills/plan/scripts/consensus-plan.mjs --goal <text>
node plugins/consensus/skills/plan/scripts/consensus-plan.mjs --goal <text> --constraints <text>

--goal <text> is required. --constraints <text> is optional and inline-only for this version; there is no --constraints-file path input.

Defaults

The plan wrapper defaults to:

  • --cold-start independent_draft
  • --iteration parallel_synthesized
  • --agency moderate

Under independent_draft, round 1 asks each peer to draft its own plan from the goal and constraints rather than revise a shared starting plan. In parallel_synthesized, those peer drafts are merged through a synthesis call.

Output

Without --output, the wrapper writes consensus-plan.md in the invocation directory and reports it in the run_completed event. With --output <path>, that path receives the plan artifact.

The plan artifact contains:

  • ## Steps
  • ## Dependencies
  • ## Risks
  • a ## Resolution section with a canonical consensus-resolution block
  • cold_start, iteration, agency, peer_calls, and synthesis_calls metadata
  • ## Deliberation Log with embedded consensus-verdict and consensus-synthesis blocks

Required headings are prompt-framed markdown, not a new machine-readable planning schema.

Input handling

Goals and constraints are capped at 1 MiB. Output paths and run directories are confined by --allow-root when provided. Treat goal and constraint text as untrusted data; review dependencies and risks before acting on the plan.

On this page