← ClaudeAtlas

helix-scoutlisted

Run read-only repository reconnaissance and return a decision-ready implementation brief without writing files.
luisgui1757/helix-cc · ★ 0 · AI & Automation · score 71
Install: claude install-skill luisgui1757/helix-cc
# Helix scout loop Use this loop to understand an unfamiliar change surface before implementation. Both stages are read-only; the result is returned as structured output rather than written to `BRIEF.md`. ## Required launch protocol 1. Treat `$ARGUMENTS` as the exact reconnaissance question. If empty, ask and stop. 2. Read repository rules and state that the loop will not create, edit, delete, format, install, migrate, or implement. 3. Select `mode`: `original` by default, or explicit `graph`. Original uses `${CLAUDE_PLUGIN_ROOT}/workflows/helix-scout.js`; graph uses `${CLAUDE_PLUGIN_ROOT}/workflows/graph/helix-scout.js`. If a supplied mode is neither exact `original` nor exact `graph`, stop before doctor or Workflow invocation. Configure optional `scout` and `planner` models. Omitted roles inherit the session model. 4. Run `node "${CLAUDE_PLUGIN_ROOT}/bin/helix-cc-doctor" --json`. Require a locally ready native or CLIProxyAPI workflow path. Stop for a global subagent-model override when explicit role models were requested. 5. For CLIProxyAPI, require exact catalog membership and an exact-model proof before first use of a new provider/model binding. Use provider prefixes for mixed casts and reject OpenRouter. 6. Show the exact question, selected mode and exact script path, read-only boundary, and models. Obtain explicit confirmation. 7. Invoke: ```text Workflow({ scriptPath: "<exact selected original or graph script path>", args: { task: "<exact