The Agent Skills Directory

[COMMAND_EXECUTION]: The skill recommends several commands ($project-spec, $project-tickets, $project-clarify) and references project-review. These are documented as user-invoked recommendations rather than automated executions. The skill itself explicitly states it 'grants nothing' and 'never edits the SPEC, tickets, code or a Charter'.
[DATA_EXPOSURE]: The skill reads from and writes to the .scratch/ directory and project metadata files (SPEC, tickets). This is consistent with its primary purpose of tracking implementation changes and proposing redirection. No network-based data exfiltration or sensitive credential access was detected.
[PROMPT_INJECTION]: Instructions emphasize following the user's explicit approval and 'never infer approval from silence, from a model's judgment, or from an earlier approval'. This serves as a defensive measure against unintended behavior changes.
[DYNAMIC_CONTEXT_INJECTION]: No use of shell injection or dynamic replacement patterns (!command) was found in the markdown files.
[REMOTE_CODE_EXECUTION]: No remote script downloads or dynamic code execution patterns were identified. All references are internal to the repository.

light-redirect