workflow-scan

Pass

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes trellis -v to resolve project paths and versions, which is a required functional step for locating the analysis target.
  • [SAFE]: The skill operates exclusively on the local filesystem within a predefined temporary directory structure. No network activity or exfiltration patterns were identified.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local project files, creating an attack surface where content from scanned files could influence the generated report.
  • Ingestion points: Local files in /tmp/trellis-{VERSION}-2/, including workflow documentation and script metadata (SKILL.md).
  • Boundary markers: Absent; the skill summarizes scanned content without using delimiters to isolate it from instructions.
  • Capability inventory: Shell execution of trellis -v and local filesystem access (SKILL.md).
  • Sanitization: Absent; content from the scanned project is directly used to populate the findings and analysis summary.
Audit Metadata
Risk Level
SAFE
Analyzed
May 20, 2026, 02:37 AM
Security Audit — agent-trust-hub — workflow-scan