scoutlisted

# scout Automatically dispatch the `scout` agent for every scout request. This skill is only a routing layer: the agent owns the read-only investigation, tool budget, evidence contract, baseline reconciliation, and `Scout Report` format. ## Route Choose one mode: | Mode | Use for | | --- | --- | | `--repo` | Repository, git, plan, or implementation context | | `--research` | Current external behavior, alternatives, or tradeoffs | | `--diagnose` | A bug, failed test, error, delta, or blast radius | | `--review` | A diff, governing intent, and risky paths | Infer the mode when omitted: default to `--repo`; use `--diagnose` for an error or failing test; use `--review` for a diff. Route unknown-unknowns or framing analysis to `blindspot` instead. ## Dispatch Call agent `scout` with: - selected mode - scope and question - known paths, patterns, changed files, or plan scope - prior evidence claims to verify, when available - a proportional tool budget - any caller-specific output fields Do not perform a parallel inline scan. Make only a single obvious lookup inline when it is cheaper than dispatching the agent. Never ask the agent to edit, design, select a final approach, or make a review verdict. Return the agent's `Scout Report` and identify the next decision it supports.