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
## Resolutionsection with a canonicalconsensus-resolutionblock cold_start,iteration,agency,peer_calls, andsynthesis_callsmetadata## Deliberation Logwith embeddedconsensus-verdictandconsensus-synthesisblocks
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.
Decide
Run the decide skill to choose between options with independent peer decision drafts, synthesis, minimal agency, explicit dissent, and resolution metadata.
Refine
Run the refine skill to deliberate two AI peers toward a converged markdown artifact, including iteration modes, resume, escalation, and host-mediated parallel sections.