creating-portable-skills

Warn

Audited by Gen Agent Trust Hub on Nov 3, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructs the agent to install evaluation-related skills from a third-party GitHub repository (github.com/ai-evals-course/evals-skills). This involves following installation instructions from an external source that is not verified as a trusted organization or well-known service.
  • [COMMAND_EXECUTION]: For the auditing process, the skill executes a bundled shell script (scripts/signal-scan.sh) to perform automated regex-based scanning of markdown files. This script uses standard utilities such as find and grep to identify specific patterns in the skill's instructions.
  • [INDIRECT_PROMPT_INJECTION]: As the skill's purpose is to audit other agent skills, it is exposed to malicious content embedded in the target packages that could influence the agent's behavior during the review.
  • Ingestion points: Reads and processes the content of SKILL.md and files within the references/ directory of any skill package being audited.
  • Boundary markers: The instructions do not specify the use of delimiters or explicit 'ignore instructions' markers when the agent extracts or reports evidence from audited files.
  • Capability inventory: The environment allows shell script execution, file system operations for creating or revising skills, and network access for external evaluation tasks.
  • Sanitization: There are no documented procedures for sanitizing or escaping the content extracted from target skill files before the agent processes it.
  • [EXTERNAL_DOWNLOADS]: The skill references external URLs for specifications (agentskills.io) and documentation (hamel.dev), in addition to the evaluation repository on GitHub.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 3, 2026, 06:19 PM