nuke-recon
Installation
SKILL.md
Nuke Recon
A small helper, not a ceremony (map: references/family-map.md): 0–1 agents, no preflight confirmation, no run_dir. It does one thing — build or refresh .nuke/repo-map.md, the durable recon cache the ceremonies otherwise produce as an audit by-product. Run it once in a new repo and every later nuke run starts warm.
Protocol
One clerk-tier scout (references/model-tiers.md) — or the orchestrator directly when spawning is unavailable — produces, per the recon rules the ceremonies share:
- Stack: manifests, languages, frameworks, workspace layout.
- Conventions: project instruction files (CLAUDE.md, AGENTS.md, .cursorrules, copilot-instructions) distilled — these define what is intentional.
- Gates: the per-path-prefix gates table per references/stack-adapters.md (test · typecheck · lint), resolved verbatim from manifest scripts → CI → fallback table; gateless areas flagged.
- Architecture sketch: top-level areas, entry points, the import direction between layers — one screen, not an essay.
- File-type → skill map: local skill library listed once, matched per references/stack-adapters.md.
- Security surfaces detected per the same file (web / CLI / library / daemon / infra / ai).
Stamp the map with date + HEAD. --refresh (or an existing stale map): verify against git history since the stamp and update deltas — never re-scan what did not change.