xy-script-glue

Pass

Audited by Gen Agent Trust Hub on Sep 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute a bundled Python utility script scripts/atoms-search.py to retrieve knowledge fragments (atoms) based on user input keywords.
  • [EXTERNAL_DOWNLOADS]: The atoms-search.py script performs HTTP POST requests to https://api.xyskill.xyz to perform remote knowledge searches. This domain is recognized as a vendor-owned resource associated with the author 'xyaz1313'.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided video scripts which could contain malicious instructions. The skill mitigates this by using specific marking protocols (strikethroughs and emojis) for changes and instructing the agent to extract only keywords for tool invocation rather than passing raw text.
  • Ingestion points: Video scripts are ingested in Phase 1 (SKILL.md).
  • Boundary markers: Phase 4 defines explicit markers like ~~original~~ and 🆕 for new text.
  • Capability inventory: Shell execution of the local Python script atoms-search.py.
  • Sanitization: Instructions require breaking user queries into 2-4 character keywords before retrieval.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 18, 2026, 05:21 AM
Security Audit — agent-trust-hub — xy-script-glue