helix-looplisted

# Helix loop Use the plugin's native Claude Code mapping of Helix `full-cycle` and guided `plan-implement` for a repository change that deserves independent planning and adversarial verification. ## Required launch protocol 1. Treat `$ARGUMENTS` as the exact task. If it is empty, ask for the task and stop. 2. Read repository rules and inspect enough of the target to identify the likely write scope and canonical verification argv vector. Do not encode shell syntax. The first item must be a PATH-resolved executable name without `/`, `\\`, or a leading dash; use `["npm", "run", "verify"]` or `["node", "scripts/check.mjs"]`, never `["./scripts/check.mjs"]`. 3. Build a requested configuration: - `mode`: `original` by default, or explicit `graph`. `original` uses `${CLAUDE_PLUGIN_ROOT}/workflows/helix-delivery.js`; `graph` uses the generated `${CLAUDE_PLUGIN_ROOT}/workflows/graph/helix-delivery.js`. If a supplied mode is neither exact `original` nor exact `graph`, stop before doctor, evidence-session, or Workflow invocation. - `maxPasses`: default `3`, allowed `1..5`. - `models.planners`: either empty or two to four model identifiers accepted by the active Claude Code backend or authenticated gateway. Empty means two planners inheriting the session model. - `models.judge`, `builder`, `tester`, `documenter`, `reviewer`, `redteam`, `verifier`: optional backend/gateway model identifiers. Omitted means the role inherits the session model. - Role ef