remediatelisted

# Remediate Drive a plan of fixes to completion without supervision, and stop honestly when a fix is beyond reach instead of faking success. The one rule that makes this safe: **green must come from fixing code, never from weakening tests.** An agent told "don't stop until green" will, when stuck, delete an assertion or add a `.skip` to force a pass. This skill forbids that structurally. You run unattended between two human gates: the user approves the *plan* up front, and triggers the *push* at the end. Everything in between is yours. ## Steps 1. **Get the plan, then gate on it.** Look for `REMEDIATION_PLAN.md` at the repo root. - If it exists, read it. Each phase is one checklist item. - If it's missing, scan the repo for real defects (failing tests, obvious bugs, error-prone spots the user named), draft a `REMEDIATION_PLAN.md` in the format below, show it to the user, and **wait for explicit approval before doing any work.** This is the only start-of-run gate; do not skip it. - Load every phase into TodoWrite, one todo per phase, so progress is visible. Plan format: ```markdown # Remediation Plan ## Phase 1: <short title> - What's wrong: <symptom / failing test / bug> - Fix: <the intended change, in source only> - Done when: <the observable condition, e.g. "test_x passes, suite green"> ``` 2. **Detect the test command once.** Auto-detect and reuse it for every phase: - `package.json` with a `test` script → `npm test` (or `pnpm tes