grep-with-context
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local utility scripts (auth.js, grep-api.js, update-check.js) using the Node.js runtime to manage authentication, file uploads, and background update checks.\n- [COMMAND_EXECUTION]: Employs
node -eone-liners to parse JSON output from API responses in a shell environment.\n- [PROMPT_INJECTION]: The skill ingests untrusted data from user-provided files (PDFs, CSVs, images) and interpolates filenames directly into research prompts, creating an attack surface for indirect prompt injection.\n - Ingestion points: Files identified via absolute paths in $ARGUMENTS or conversation context (SKILL.md Step 1).\n
- Boundary markers: The skill instructs the agent to explicitly mention filenames in the refined prompt (Step 4), which helps delimit context but does not fully neutralize adversarial inputs.\n
- Capability inventory: Includes file reading ([ -f $FILE ]), network upload via grep-api.js, and submission of research jobs via the same script.\n
- Sanitization: There is no evidence of filename sanitization or content filtering before the data is passed to the research engine.
Audit Metadata