The Agent Skills Directory

[SAFE]: The skill performs standard local file management within a defined project hierarchy (specifically 'state/tasks/' and 'state/lessonbook/'). No unauthorized file system access, network operations, or credential exposure patterns were detected.- [PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it processes user feedback and external inbox files to influence the agent's active constraints. Ingestion points: 'state/lessonbook/inbox.md' and user corrections. Boundary markers: The skill includes strong negative constraints stating the agent 'never gains authority' and requires a 'human review gate' for all entries. Capability inventory: Appending to 'correction_journal.md' and updating 'run_state.yaml'. Sanitization: Data is structured using a mandatory markdown template. The risk is considered safe as the behavior is the primary intended function of the skill and includes appropriate human-in-the-loop safeguards.

correction-capture