discovery-plan

Run the discovery-plan skill after research synthesis: it reads the context.md ledger, research runs, and active jobs, then writes research-plan.md with cited evidence, open questions, assumptions with confidence and disproof, and technical research with integration costs.

ateam-plan.md holds what the team will do: job-traced goals, deliverables to reach v0, initiatives with owners and dates, and decision criteria. research-plan.md holds what is unknown or assumed: open questions, assumptions with confidence, and technical research. Both are written together so they cannot drift.

Use discovery-plan when unvalidated assumptions or open questions still block confident scope decisions. Once requirements are validated and delivery is underway, use prd-writer instead; discovery-plan hands off to wireflow, page-brief, then prd-writer.

No durable writes occur without a human. The read-back gate requires review, so with nobody present the skill delivers drafts in its report only, writes nothing, and states the run awaits review. An unreviewed research plan is never committed.

project-plan.md is written only at the pr phase's plan refresh, once a v0 exists and is integrated. It covers what comes after v0 for the human team. On every other invocation it is omitted entirely, since planning past a nonexistent v0 is fiction.

It never synthesizes raw evidence (research-synthesis must run first) and never creates or rewrites job statements (jobs-to-be-done owns those). Plans built on unsynthesized evidence plan the wrong unknowns, so the skill stops and routes upstream.