Skills
skills/yigitkonur/skills-by-yigitkonur/enhance-skill-by-derailment/Gen Agent Trust Hub

enhance-skill-by-derailment

Warn

Audited by Gen Agent Trust Hub on May 10, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches skill definitions and reference materials from user-specified GitHub repositories using the GitHub CLI (gh api).
  • [REMOTE_CODE_EXECUTION]: Directly executes remote instructions by downloading them from external sources and passing them to a subagent for execution. This enables unvetted external content to control agent behavior within the test environment.
  • [COMMAND_EXECUTION]: Employs several shell commands (including gh, find, grep, and wc) and a Python one-liner to download files, extract execution traces from JSONL files, and verify the integrity of the skill structure.
  • [DATA_EXFILTRATION]: By executing untrusted remote instructions, the skill creates an attack surface where a malicious skill could instruct the subagent to exfiltrate sensitive data from the execution environment.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 10, 2026, 03:48 PM