The Agent Skills Directory

[SAFE]: The skill enforces a human-in-the-loop requirement for any changes to permanent configurations. It explicitly prohibits the agent from writing to the control/ directory or modifying rules.
[SAFE]: No network access, credential handling, or external code execution patterns were detected.
[INDIRECT_PROMPT_INJECTION]: The skill processes correction_journal.md, which contains logs from previous tasks. While this data could theoretically contain malicious patterns, the skill mitigates this by adding not_authoritative_until_reviewed: true to its output and restricting its actions to writing proposals for human review.

lesson-propose