Feedback Learning V2

What to check before installing: 1) Inspect SKILL.md for hidden characters: the pre-scan flagged unicode control characters — open the file in a hex or 'show-nonprinting' view to ensure nothing is obfuscated. 2) Review and audit the scripts: they are pure Python/shell and do not contact external hosts, but you should manually read them (especially log-event.sh, error-catcher.sh, and detect-feedback.py) to confirm no unexpected behavior. 3) Consider sensitive-data exposure: the optional PostToolUse hook and log-event calls capture TOOL_COMMAND and TOOL_STDERR and write them into events.jsonl and reports. If your tools might print secrets (tokens, passwords, database URLs, stack traces containing keys), that data will be stored locally. If you enable this skill, either filter/redact such output or restrict the directory's filesystem permissions and retention. 4) Add hooks/crons intentionally: the skill requires manual changes to AGENTS.md and cron entries; do not enable these automatically. Only enable the hook if you accept ongoing local logging and automated promotion of rules. 5) Backup/retention: ensure events.jsonl and reports are stored where you control retention and access. Remove or rotate logs if they accumulate sensitive content. 6) If you are unsure, run the scripts in a sandboxed account or container first to observe what gets captured before enabling them in your production agent environment.