Apply Learnings

Before installing or running this skill, be aware it will scan your conversation/session transcripts and tool inputs and can write extracted "learnings" into global files (like ~/.claude/CLAUDE.md or ~/.claude/MEMORY.md) and into other skills' reference directories. That process can capture and persist sensitive data (passwords, API keys, secrets pasted in sessions, or other PII). Recommended steps: - Inspect the full scripts locally (scripts/analyze_session.py) to verify exactly which paths are read and written and whether there is any network I/O. Confirm there are explicit redaction measures for secrets. - Run initially with the narrowest scope (e.g., --scope current) and review all extracted learnings before allowing any writes. - Back up the target files/directories you allow the skill to write (project CLAUDE.md, ~/.claude/*, and ~/.claude/skills/*) so you can revert unintended changes. - Deny or carefully review suggestions to persist learnings to shared/global files or to append into other skills' directories; prefer storing only project-local, user-reviewed notes. - If you use cross-machine syncing (TerraBlob or similar), consider disabling it while testing so learnings (which may include sensitive context) are not propagated automatically. - If you are unsure about whether the script modifies other skills or performs network access, run it in an isolated environment (e.g., a throwaway account or VM) first and monitor filesystem changes. If you want, I can (1) scan the rest of the script for any network calls or explicit file-write code you should watch for, or (2) suggest a safe reduced workflow for extracting and redacting learnings before persistence.