parallel-api

Pass

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external web content (search results, page excerpts, and research outputs), which inherently creates a surface for indirect prompt injection.
  • Ingestion points: The agent retrieves data from external sources via Search, Extract, and Responses API endpoints as described in the Patterns section of SKILL.md.
  • Boundary markers: The skill includes a specific mitigation rule (Rule 7) that explicitly instructs the agent to treat all returned web content as untrusted data and to never follow instructions found within excerpts, pages, or research output.
  • Capability inventory: The skill allows the agent to perform network requests to vendor APIs, write to the filesystem (e.g., creating .env files), and execute shell commands for environment validation.
  • Sanitization: While the instructions mandate a defensive posture toward untrusted data, no automated sanitization or structural validation of the retrieved text is detailed.
  • [EXTERNAL_DOWNLOADS]: The skill guides the agent to install necessary SDKs and libraries for integration.
  • Evidence: The instructions include commands for pip install parallel-web, npm install parallel-web, pip install requests, pip install openai, and npm install svix.
  • Context: These installations involve either well-known industry standard libraries or the official SDK provided by the skill's author (parallel-web), which is considered a legitimate vendor resource.
  • [COMMAND_EXECUTION]: The skill provides utility shell commands to assist with project setup and verification.
  • Evidence: SKILL.md contains bash snippets for checking the presence of environment variables (e.g., [ -n "$PARALLEL_API_KEY" ]) and validating files (e.g., grep -qs '^PARALLEL_API_KEY=' .env).
  • Verification: The instructions include a manual verification step requiring the execution of a test request to confirm the API integration is functional.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 12, 2026, 12:35 AM