The Agent Skills Directory

[SAFE]: No malicious patterns, data exfiltration, or unauthorized command execution detected. The skill is designed to manage personal productivity data within a local markdown vault environment.
[PROMPT_INJECTION]: The instructions focus on establishing safety boundaries and user-consent workflows. Specifically, the 'Iron Law' and 'Role' definitions explicitly prevent the AI from making unilateral changes to project state or performing work outside the defined management scope.
[DATA_EXFILTRATION]: The skill operates on a markdown vault and uses a configuration system (Personal OS config) for path resolution. No network operations, hardcoded credentials, or exfiltration patterns to external domains were identified.
[COMMAND_EXECUTION]: No dangerous shell commands, subprocess spawning, or remote code execution patterns were identified. The mentioned 'sync-workspaces' is described as an advisory disk report rather than an automated execution vector.
[INDIRECT_PROMPT_INJECTION]: The skill processes project status and inbox data from a vault. While it ingests untrusted user data, its capabilities are restricted to metadata auditing and note generation. The requirement for explicit user confirmation ('PROPOSED' status) serves as a robust control against unintended state changes driven by injected content.

review-week