platform-custom-metadata-type-generate

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data to generate metadata records, representing a potential injection surface. \n
  • Ingestion points: User-supplied labels, CSV files, and prose descriptions are used to populate record values and filenames as described in Section 4. \n
  • Boundary markers: None provided for the raw user-supplied data. \n
  • Capability inventory: The agent executes a local shell script (scripts/sanitize-developer-name.sh) and writes multiple XML metadata files to the filesystem. \n
  • Sanitization: The skill implements identifier sanitization through a shell script that uses strict regex to permit only alphanumeric characters and single underscores. \n- [COMMAND_EXECUTION]: The skill requires the agent to run a bundled shell script, scripts/sanitize-developer-name.sh, for string processing. The script is restricted to basic string manipulation using sed and cut, uses safe quoting for arguments, and performs no network or high-privilege operations. \n- [SAFE]: The skill demonstrates a strong security posture by instructing the agent to warn users against storing secrets or API keys in metadata, instead recommending Named Credentials. It also enforces a 'no-silent-downgrade' policy for visibility settings, ensuring that components intended to be 'Protected' are not accidentally exposed as 'Public' in production-like environments.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 09:54 AM
Security Audit — agent-trust-hub — platform-custom-metadata-type-generate