captains-boxlisted

# captains-box — writing and validating gated plans A build-authorizing plan is a dispatch release: it may not authorize anything until every claim it stands on is classified, cited, and checkable. The canonical grammar lives in the header of `plan-gate.mjs` — consult it before writing a plan. This skill is the workflow around it. ## Workflow 1. **Recon before design.** Run `node plan-gate.mjs --recon <paths the plan will touch>` and paste the output into the plan's `## RECON PACK` section. Read the SETTLED DECISIONS it surfaces — these are settled rulings; list them under `## PRESERVE`. 2. **Write the plan** with frontmatter `authorizes-build: true` and the sections the grammar requires: RECON PACK, TOUCHES, PRESERVE, PREMISE LEDGER, CAPTAIN'S BOX, SIMPLEST-PATH. 3. **Ledger discipline.** For each load-bearing claim: state FACT and INFERENCE separately; pick the CLASS honestly (a claim about an external system needs a captured fixture — code only shows your expectation, never the server); quote real lines for `path:line` evidence — the gate byte-checks them. 4. **Evidence is real or the claim is ASSUMED.** A claim that cannot be verified is labeled `FACT [ASSUMED]` with a DISPOSITION, and surfaces as an OPEN QUESTION for the risk owner. An honest ASSUMED passes review; a fabricated VERIFIED fails the byte-check. 5. **Validate:** `node plan-gate.mjs <plan.md>` (review mode). Fix every FAIL. Present remaining OPEN questions to the huma