Skill Recon

Recon

Map unfamiliar or uncertain territory before committing effort.

Recon checks whether a request describes the world that actually exists. It has three modes: clarify a fuzzy brief, map the decision dependencies of a large initiative, or examine an external project that overlaps existing work.

Use it when: Initial reads leave a material uncertainty about the target, hidden coupling, missing decisions, or what could be reused before adoption.

Use something simpler when: A couple of ordinary reads settle the request, the target and check are explicit, or the only problem is unfamiliarity.

Illustrative request

Before replacing our import pipeline with this library, inspect the overlap, what we could reuse, and which adoption questions remain unresolved.

What a useful result looks like: A corrected brief, a decision-dependency map, or a bounded adoption/harvest record with evidence and residual questions. The map helps the task owner choose the next authorized action.

Boundary: Recon is read-only and ends at understanding. It neither runs an untrusted candidate nor decides adoption. Finding a possible repair does not make that repair part of the recon assignment.

Resolve settles an evidence question; Open Questions helps the user decide consequential remaining forks.

Read the canonical v7.3.0 method · All methods