| name | wipnote:arch-bootstrap |
| description | Bootstrap architectural memory for a repo — generates starter cards from a structured brief |
Architectural Memory Bootstrap
Use this skill to populate a project's architectural memory from scratch. It runs the bootstrap command, reads the brief, and authors starter arch cards via the validated add path.
Trigger keywords: bootstrap arch, bootstrap architectural memory, starter cards, arch bootstrap, initialize arch cards
Step 1: Generate the Brief
wipnote arch bootstrap
Read the full output. It contains:
- Repo Layout — top-level directories and Go module names
- Existing Docs — CLAUDE.md / AGENTS.md content (instructions, not architecture)
- Lineage Hotspots — files touched most in the last 90 days (signals where arch cards are most valuable)
- Authoring Instructions — card kinds, constraints, command template
Step 2: Author Cards
For each major subsystem, invariant, or hazard you identify from the brief, run:
wipnote arch add <slug> \
--kind <kind> \
--created-by agent \
--paths "<glob1>,<glob2>" \
--body "<body up to 120 words>"
Card kinds:
subsystem-map — what a subsystem does and its boundaries
invariant — a constraint that must always hold
hazard — a known failure mode or sharp edge
decision — a recorded architectural decision
Rules:
- Body: max 120 words
- Slug: lowercase letters, digits, hyphens only
- Globs: use
** for recursive match (e.g. internal/**, cmd/wipnote/*.go)
- Name and path-glob set must each be unique across all cards
Aim for 3–8 cards covering the top subsystems visible in the hotspot list.
Do NOT mechanically import CLAUDE.md content — those are instructions, not architecture. Focus on structural facts: what systems exist, what invariants must hold, what hazards to avoid.
Step 3: Validate
wipnote arch validate
Fix any reported errors (body too long, missing required fields, invalid kind).
Step 4: Review
wipnote arch list
wipnote arch show <slug>
Verify the cards describe real structural facts about the codebase, not process instructions.