best-practices

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by fetching industry best practices and standards from the public internet using the WebSearch and WebFetch tools.
  • Ingestion points: External data from web searches and direct URL fetches are ingested into the agent context in Phase 1 (Industry Research) as documented in SKILL.md.
  • Boundary markers: The skill includes an "Untrusted content rule" explicitly instructing the agent to treat all fetched content as untrusted, ignore instructions found in pages, and cross-verify claims across independent sources.
  • Capability inventory: The skill has access to the filesystem (Read, Grep, Glob) and can orchestrate sub-agents and other skills (Agent, Skill).
  • Sanitization: The instructions specifically forbid executing commands or code snippets retrieved from external web sources.
  • [COMMAND_EXECUTION]: The skill performs repository-wide searches and file reads based on a user-provided scope.
  • The instructions include a mandatory validation step for the --scope argument, requiring the rejection of absolute paths and parent directory traversal (..) to prevent unauthorized access to the host filesystem.
  • [EXTERNAL_DOWNLOADS]: The workflow relies on retrieving external documentation, RFCs, and industry standards.
  • The skill uses a prioritized cascade of web tools (agent-browser, WebSearch, WebFetch) to gather research data from the internet.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 03:38 AM
Security Audit — agent-trust-hub — best-practices