action-runner

Pass

Audited by Gen Agent Trust Hub on Jun 25, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local Python scripts scripts/action_preview.py and ../professional-brain/scripts/brain_write.py to classify risks and log execution results to a local tracking system.
  • [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) because it ingests and acts upon untrusted data from external recommendations or checklists.
  • Ingestion points: Processes external launch checklists, PRD requirements, and postmortem follow-ups as defined in SKILL.md.
  • Boundary markers: The instructions lack explicit delimitation or instructions for the agent to ignore control sequences embedded within the ingested recommendations.
  • Capability inventory: The skill has the capability to perform outbound and destructive operations (delete, post, deploy, charge) via connected MCP tools like GitHub and Slack.
  • Sanitization: Risk classification in scripts/action_preview.py relies on a fixed regex of operation keywords; it does not validate the content of the arguments and could potentially be bypassed by using synonyms (e.g., 'wipe' or 'terminate') not included in the high-risk pattern.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 25, 2026, 09:21 AM
Security Audit — agent-trust-hub — action-runner