soia-meta-sync-skills
Pass
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The Python script sync_soia_skills.py performs file system operations including creating directories, deleting files or directories, and creating symbolic links. These operations are localized to specific agent skill directories and are gated by a dry-run mode and explicit user confirmation. The script is designed to manage specific skill folders and does not access sensitive system paths.
- [DATA_EXFILTRATION]: The skill maintains local audit logs in the user's state directory. The implementation explicitly avoids recording sensitive information or credentials, and the logs are not transmitted externally.
- [SAFE]: The skill references external resources for installation and documentation, all of which point to the author's own repositories on GitHub or well-known service providers. The included Python script does not perform any network operations.
- [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection in scripts/sync_soia_skills.py where it parses metadata from other skills. 1. Ingestion points: Metadata is read from SKILL.md files in the source directory via the parse_hard_deps function. 2. Boundary markers: No delimiters or safety warnings are used when reading these files. 3. Capability inventory: The script can perform file system modifications such as shutil.rmtree and symlink_to. 4. Sanitization: The script performs minimal cleaning of input tokens from the metadata files. However, the risk is mitigated because the script only operates on directories that already exist within the provided source path.
Audit Metadata