What problem does it solve? Reading large source surfaces (multiple files, packages, or folders) directly floods an agent's context with noise. This Skill primes the calling agent with high-signal context by dispatching parallel sub-agents that absorb the noise and return only digested findings with verified source:line citations. ## Core Features & Use Cases - Parallel delegated investigation: Spawns up to 3 sub-agents per wave across single-source, multi-source, discovery, or mixed input modes, keeping the caller's context clean. - Proof-of-work verification: Every finding and unfollowed pointer requires a source:line plus verbatim quote, verified via Grep against files or temp-file-backed dumps — never from memory. - Iterative wave ladder with sufficiency gate: Runs up to 8 waves (silent ≤3, logged 4-6, user approval at 7) and declares primed, partial, or stuck based on an explicit checklist. - Use Case: You receive a handoff listing five unfamiliar packages to modify. Invoke /prime-sweep to get a primed summary, a source:line index for first-hand re-reading, discarded-noise categories, and known gaps — without reading any source yourself. ## Quick Start Invoke /prime-sweep with the list of source files or folders you need primed context from before starting implementation work.
