regulatory-compliance

Pass

Audited by Gen Agent Trust Hub on Mar 23, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection (Category 8) due to its core function of auditing untrusted codebase files.
  • Ingestion points: The agent reads various configuration files (package.json, requirements.txt, go.mod, etc.) and scans source code across the project.
  • Boundary markers: No delimiters or specific instructions are provided to distinguish between the agent's instructions and the data being audited.
  • Capability inventory: The skill has broad filesystem read access and the ability to write local telemetry logs to ~/.claude/projects/.
  • Sanitization: No sanitization or escaping of ingested content is performed before processing.
  • [PROMPT_INJECTION]: The skill includes a persistence mechanism through its 'SELF-EVOLUTION TELEMETRY' section, which instructs the agent to append metadata to a local file (~/.claude/projects/skill-telemetry.md) to track state and performance across executions.
  • [DATA_EXPOSURE]: As part of its auditing purpose, the skill identifies high-value security targets including encryption implementations (bcrypt, AES libraries), KMS integrations, and PII/PHI storage locations, which could be exploited if the agent's context is compromised.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 23, 2026, 10:59 AM
Security Audit — agent-trust-hub — regulatory-compliance