lab-automation

Pass

Audited by Gen Agent Trust Hub on Mar 23, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill uses the $ARGUMENTS variable to allow users to specify a focus area for the audit. This creates a surface where untrusted data could potentially override agent instructions.
  • Ingestion points: SKILL.md (SCOPE: $ARGUMENTS)
  • Boundary markers: Absent. The input is interpolated directly without delimiters or safety warnings.
  • Capability inventory: The skill has the ability to write files to the local directory (docs/lab-automation-analysis.md) and the user's home directory (~/.claude/projects/).
  • Sanitization: No validation or escaping of input data is performed before processing.
  • [DATA_EXFILTRATION]: The skill implements a 'Self-Evolution Telemetry' feature that writes execution metadata to ~/.claude/projects/skill-telemetry.md. This is a local data logging behavior and does not involve transmitting data to a remote server.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 23, 2026, 10:57 AM
Security Audit — agent-trust-hub — lab-automation