scout

# Scout Token-efficient parallel scouting that finds the right files before you touch them. Tuned for multi-discipline repos: app code, data pipelines, IaC, and BI artifacts. ## Modes | Argument | What | When | |---|---|---| | _(none)_ | **Internal** - Explore subagents in parallel (`references/internal-scouting.md`) | Default in Claude Code. Best for ≥6 logical segments. | | `ext` | **External** - Gemini / OpenCode CLI in parallel (`references/external-scouting.md`) | Large surfaces (1M+ ctx) and you have one of the CLIs installed. SCALE 1–5. | Internal mode needs the Task/Explore tool (Claude Code). In a runtime without it (e.g. Codex), use `ext` mode - run its `gemini`/`opencode` bash commands sequentially - or fall back to inline `Glob`/`Grep` sweeps. ## When to use - About to change something that **could span multiple folders** - e.g. add a new dbt source that touches `models/`, `schema.yml`, `dashboards/`, and a CI workflow - User says **"find / locate / search for"** - code, models, charts, secrets, manifests - Starting a debug session and need a file map before invoking `vd:debug` - Before a refactor, migration, or deletion that could ripple across services - Auditing a repo you don't own well - what's where, and how does it wire together **Skip scout when** the target file is already known, or one `grep` answers the question. Scout is for breadth, not pinpoint lookups. ## Quick start 1. Parse the prompt → list **search targets** (names, patterns, behaviors)...