root-cause-before-fixeslisted

# Root cause before fixes **Iron Law: no fixes without root-cause investigation first.** A fix applied to a symptom costs three later sessions; five minutes of tracing costs five minutes. ## Phases (in order, no skipping) 1. **Reproduce and READ the actual error.** The real message, the real exit code, the real line — not your memory of similar errors. On this repo the transcript keeps every failure; quote it to yourself before theorizing. 2. **Pull what's known.** `py agent_cli.py recall-at --path <file>` (or `--command "..."`) surfaces prior lessons and anti-patterns for this exact target — someone may have paid for this bug already. (Hooks do this automatically before Edit/shell actions; pull manually when investigating.) 3. **Trace to the cause, not the site.** Compare against the nearest WORKING analog (a passing test, a sibling module) and diff the assumptions. 4. **State a single hypothesis out loud:** "I think X is the root cause because Y." If you cannot say Y, you are not done investigating. 5. **Fix with evidence.** Failing test (or failing command) first, watch it fail, apply the fix, watch it pass. An Edit that "should" fix it is a hypothesis, not a fix. ## Red-flag rationalizations (stop when you hear yourself think these) - "Just try this first, then investigate if it doesn't work" - "I don't fully understand it, but this might work" - "It's probably the same issue as before" (verify — this repo's history rewrote SHAs once) - "The er