defer-finding
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect prompt injection surface identified. The skill ingests data from 'docs/findings.json' and uses it to generate issue titles and bodies. \n- Ingestion points: 'docs/findings.json' (Step 1). \n- Boundary markers: Absent; no delimiters separate untrusted finding content from agent instructions. \n- Capability inventory: Shell command execution via 'gh' CLI and file system access. \n- Sanitization: Absent; the agent is instructed to rewrite and distill potentially malicious content. \n- [COMMAND_EXECUTION]: Potential command injection surface in Step 2. The skill interpolates keywords derived from finding titles directly into a 'gh issue list --search' command. If a finding title contains shell metacharacters, it could lead to unintended command execution.
Audit Metadata