Strip console warnings before classifying infra failures

[hiroshinishio]

Summary

• detect_infra_failure was substring-matching AccessDeniedException in app-level console.warn output (SSM fetch fallback), sending PRs with real Jest failures down the infra-retry branch. Three empty-commit retries later, GA gave up without ever invoking the LLM fix path.
• Fix: call strip_jest_noise first so console.warn/console.log blocks (including the orphaned AWS stack traces and closing braces) are removed before pattern scanning. The classifier's single responsibility — match infra patterns in cleaned log — is preserved.
• Regression test loads the real 396KB CircleCI log captured from Foxquilt/foxden-rating-quoting-backend PR 714 and asserts it now classifies as None.

Notes

• CLAUDE.md: documented that .env and .venv are already sourced before Claude starts, plus the python -m pytest ... invocation (so top-level imports like from constants.models ... resolve without PYTHONPATH).
• .gitignore: .claude/scheduled_tasks.lock.

Social Media Post (GitAuto)

Infrastructure-failure classifier no longer trips on app-level console warnings

• Console.warn/log blocks stripped before scanning, so app warnings (failed SSM fetches, etc.) can't false-positive the classifier
• Regression test uses a real captured CircleCI log that previously burned three empty-commit retries
• PRs with real test failures now reach the LLM fix path on the first run

Social Media Post (Wes)

Spent the morning on a classifier that confidently called a code bug "infra" because the app logged an AccessDenied warning during test setup. Three empty-commit retries later, nothing fixed. Strip the console blocks before scanning and the same log classifies correctly. One responsibility per function, clean input in.