v1.0.0 — Compressed Plans, Generated State, Evidence-Gated Completion and Spec-Driven Planning for AI Coding Agents

Plannable v1.0.0

First stable release. Plannable turns a product idea into a short, human-readable MASTER_PLAN.md plus compressed, agent-readable PlannablePlan part files — so AI coding agents work one phase at a time instead of drowning in a giant Markdown plan. Think spec-kit, but the plan is split into compressed parts where each part knows its place in the masterplan.

What Changed

• Added a CTX: block to every part file — product goal, what prior parts delivered, and what comes next — so an agent loading one part gets the whole masterplan arc without reading the others.
• Made PLAN_STATE.md a generated view: MASTER_PLAN.md checkboxes and PLAN_EVIDENCE.md are the source of truth, so plan state can no longer silently drift.
• Gated completion on evidence — plannable complete refuses to check off a part until proof is recorded.
• Taught plannable verify to warn when a plan still contains generic draft wording that hasn't been enriched for the real product.
• Improved plannable compress to map task/acceptance/verification/context sections into the right blocks and report estimated token savings.
• Grouped status and run-next output by masterplan phase.
• Added friendly "No Plannable plan found here" errors, a plannable --version flag, and parser tolerance for hand-written plans (CRLF, spacing around =).
• Froze the @PlannablePlan v0.1 format with a documented stability and versioning policy.
• Shipped skill instructions for Claude Code, Codex, and Cursor, plus a condensed README and finalized npm metadata.

Why It Matters

Plannable now gives agents dense, self-contained context per phase and makes "done" mean something: every checkoff is backed by recorded evidence, and the plan files validate structurally. Plans stay short for humans and progressive for agents, with no state drift between the files.

Reminder

Run plannable create "<product idea>" to generate a plan, enrich the draft with product-specific detail, then run-next → implement → evidence → complete → verify. Agents should load only the next pending part — its CTX: block carries the rest.