privileged-access

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill contains the phrase 'ignore previous instructions', but this is used in a defensive context to teach the agent to identify and flag prompt injection attempts within the audited data rather than executing them.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data such as session logs and vault metadata. It implements a 'Security Boundary' and 'Injection Hardening' section that explicitly instructs the agent to treat all input as untrusted and forbids the execution of embedded instructions.
  • Ingestion points: PAM configuration files, vault metadata, and session logs (via Read/Grep tools).
  • Boundary markers: Present; defined in 'Injection Hardening' and 'Prompt Injection Safety Notice' sections.
  • Capability inventory: Limited to file read and search tools (Read, Grep, Glob); no network or write capabilities.
  • Sanitization: Instructions specify that suspicious input patterns should be reported as findings rather than followed.
  • [DATA_EXPOSURE]: While the skill mentions sensitive paths like ~/.aws/credentials and .env files, these are listed as examples for inventory and audit purposes. The skill explicitly forbids the exfiltration of any credentials discovered during the review.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 01:48 PM
Security Audit — agent-trust-hub — privileged-access