agent-code-assistant

Pass

Audited by Gen Agent Trust Hub on Jun 21, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill manifest defines a robust least-privilege security model, explicitly limiting the agent's permissions to read-only access for file system and network operations.
  • [SAFE]: Secrets management is handled correctly by using environment variable placeholders for the AGENTVERSE_TOKEN rather than hardcoding sensitive credentials.
  • [SAFE]: All external resources, including the required 'agentverse' binary and the associated GitHub repository, are verified vendor resources belonging to the skill's author context.
  • [PROMPT_INJECTION]: The skill contains an indirect prompt injection attack surface as it is designed to review and process untrusted external code and pull request data.
  • Ingestion points: Source code diffs and PR descriptions processed by the 'review_code' and 'write_release_notes' skills.
  • Boundary markers: No explicit delimiters or instructional boundaries are defined to distinguish untrusted data from the agent's core logic.
  • Capability inventory: The agent is restricted to read-only filesystem and network access, preventing high-risk actions such as unauthorized file modification or code execution.
  • Sanitization: No sanitization or validation logic for ingested code content is specified in the manifest.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 21, 2026, 07:53 AM
Security Audit — agent-trust-hub — agent-code-assistant