agent-skills-evaluator

Pass

Audited by Gen Agent Trust Hub on Mar 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill requires the execution of the agtm command-line interface to perform setup, logging, and evaluation tasks. This includes commands like agtm setup, agtm skills log, and agtm skills rate which interact with the local filesystem and system environment.
  • [EXTERNAL_DOWNLOADS]: The documentation indicates that the Agtm Skills CLI has the capability to download skill bundles directly from GitHub repositories to manage agent extensions.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection during the evaluation phase because it processes untrusted data from previous agent runs.
  • Ingestion points: Task inputs and outputs are read from log files stored in .agtm/skills/log/ and processed by the agtm skills rate prepare command.
  • Boundary markers: The provided example does not show explicit delimiters or "ignore embedded instructions" warnings in the evaluation prompt template.
  • Capability inventory: The skill performs filesystem writes to save logs and benchmark configurations, and it generates output tables in the terminal.
  • Sanitization: There is no evidence of sanitization or validation of the input/output data before it is interpolated into the evaluation prompt.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 20, 2026, 10:05 AM
Security Audit — agent-trust-hub — agent-skills-evaluator