/prime-sweep

Prime the calling agent with high-signal context from large source surfaces by dispatching parallel delegated sub-agents. The calling agent never reads the sources — sub-agents absorb the noise, return only digested findings + source:line pointers. Full proof-of-work verification on every finding.

Structured execution spec: protocol.yon. Read it for the canonical rules and step sequence; this file is explanation. The two must stay in sync — if you edit one, update the other and refresh the @STAMP date.

Caller Options. Before executing, run the Caller Options protocol (caller-options/SKILL.md). Modes here are the input categories — single-source / multi-source / discovery / mixed — which shape how sub-agents are sliced.

Family invariant — /prime-* skills delegate. Every /prime-* skill uses sub-agents to keep the caller's context clean. /prime-fetch and /prime-sweep additionally require proof-of-work on every finding (source:line + verbatim quote ≥8 words OR a verbatim-token). /prime-expand is exempt — it produces intent, not findings. See /prime-fetch for the verbatim-token definition.

Boundary

Use /prime-sweep when the source surface is too large to read directly (threshold: ≥3 distinct sources OR ≥~10k tokens) and you need primed context from it. Do NOT use it for a single noisy tool call (use /prime-fetch), a vague gap that needs articulating first (use /prime-expand), or work execution (use /orchestrate-mode or /multi-agent-mode). Do NOT use it for post-hoc review (use /cold-review).

Step 1 — Trigger routing + gate

Identify the trigger: