visual-planlisted
Install: claude install-skill scoobydrew83/skills
# Visual Plan
Planning as a reviewable surface, not a wall of markdown. Inspired by
agent-native's `/visual-plan`, inverted to fit the Conductor Method: their plan
is a hosted document that can drift from the code; ours is a **lens over the
repo's existing ground truth** plus one small authored file, all tracked, all
lint-able, all blockable. Full analysis and rationale: [reference.md](./reference.md).
## The artifacts (all under the target repo)
| File | Who writes it | Role |
|---|---|---|
| `.harness/plan/blocks.json` | planner (you) | The authored plan: typed blocks, schema v2 |
| `.harness/plan/comments.jsonl` | human (page/CLI) + you (resolutions) | The review channel — comments are BLOCKED-criteria |
| `FEATURES.json` | planner commit | Acceptance criteria the plan promises; verifier grades these |
| `.harness/plan/plan.html` | `tools/render-plan.mjs` | DERIVED. Never hand-edit — same sin as editing `generated/*` |
## Block schema v2 (the contract)
Every block: stable `id` (comments pin to it), `type`, `section`, `title`,
`features` (ids that must exist in FEATURES.json). Types: `note`, `diagram`
(mermaid source), `wireframe` (ASCII sketch), `decision` (choice + rationale +
rejected — rationale mandatory), `annotated-code` (file + per-line notes; the
renderer pulls live source, so wrong lines are visible), `question` (needs
`answer` or it blocks its features).
## The rituals
**Planning (`commands/vplan.md`):** investigate → write blocks + FEATURES
entries → `nod