recon-swarmlisted
Install: claude install-skill hjr15/claude-kit
# /recon-swarm — Adaptive Read-Only Scout Swarm
## Overview
Reconnaissance before planning: fan out `Explore` agents across the codebase under independent lenses, then synthesize findings into a recon brief. The brief gives [[grill-brainstorm-build]] (and solve) something to interrogate against reality rather than assumptions.
This skill only reads. It produces no changes.
## When to Use
- Non-trivial work item or idea where affected surfaces are unclear
- Before invoking [[grill-brainstorm-build]] on anything that touches more than one file or system boundary
- Standalone "map X before we plan it" requests
- When you suspect a plan-from-memory risk (drift, hidden dependencies, unknown owners)
## Adaptive Lens Selection
Pick lenses for the job — more lenses for larger scope, fewer for narrow changes. The menu below is a starting set; ad-hoc lenses are allowed; **no upper limit**.
| Lens | What the scout maps |
|---|---|
| `runtime/ops` | How the component runs in production: startup, config, health, restarts, logs |
| `data-layer` | Schemas, migrations, persistence contracts, query patterns |
| `frontend/UI` | Components, routes, state, visual contracts affected by the change |
| `risk/security` | Auth boundaries, secret handling, injection surfaces, trust assumptions |
| `integration/contract` | API boundaries, event contracts, upstream/downstream dependencies |
| `prior-art` | Existing implementations, adjacent patterns, past decisions, test coverage |
`prior-art` ru