research-url

Pass

Audited by Gen Agent Trust Hub on Aug 16, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill has an indirect prompt injection surface because it fetches and processes untrusted data from user-provided external URLs (e.g., GitHub repositories, Gists, and websites).
  • Ingestion points: The subagent fetches content from user-provided URLs and GitHub APIs as specified in the 'Brief template' section of SKILL.md.
  • Boundary markers: The distilled report template includes a mandatory provenance line: 'Provenance: REPORTED. Distilled from the source; claims are the author's, not independently verified.'
  • Capability inventory: The skill writes distilled reports to the file system at ~/masumi/agentsmith/research/ and appends extracted rules to .superstack/learnings.md.
  • Sanitization: No explicit programmatic sanitization or escaping is performed on the ingested content; the skill relies on the subagent's distillation process to filter the data.
  • [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch external content for research purposes.
  • Evidence: The subagent is instructed to 'Fetch the source (raw URL for GitHub/gists; enumerate the full file tree via the API for repos and read EVERY substantive content file...)' in SKILL.md.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 16, 2026, 01:31 PM
Security Audit — agent-trust-hub — research-url