define-project

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONCREDENTIALS_UNSAFE
Full Analysis
  • [SAFE]: Comprehensive analysis of the skill's instructions and templates revealed no malicious patterns or security risks.
  • [INDIRECT_PROMPT_INJECTION]: The skill identifies repository content as a trust boundary and explicitly instructs the agent to ignore any embedded directives or commands found within scanned files, effectively mitigating indirect prompt injection risks.
  • [DATA_EXFILTRATION]: The brownfield scan process is restricted by a 2 MiB read budget and includes a mandatory redaction pass to identify and strip sensitive information such as private keys and credentials from the data before it is processed.
  • [CREDENTIALS_UNSAFE]: An example database connection string identified by static analysis in brownfield-scan.md was verified to be a benign placeholder used solely for documenting the skill's redaction logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 09:38 AM
Security Audit — agent-trust-hub — define-project