lab-setup

Warn

Audited by Gen Agent Trust Hub on Jul 23, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill downloads a JSON manifest from 'agency-lab.glebkalinin.com' and is instructed to execute the shell commands found in the 'install' field verbatim. This allows a remote service to dictate which commands are executed on the host system if the manifest is manipulated.
  • [COMMAND_EXECUTION]: The skill performs environment discovery by running several shell commands to check for existing software and global npm packages.
  • [DATA_EXFILTRATION]: The skill reads from local configuration files to determine the user's environment and includes this information in requests to a remote API.
  • Evidence: Reads the 'cohort' slug from '/.config.md' to construct the API URL for the curriculum manifest.
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by processing untrusted remote data that directly influences agent execution capabilities.
  • Ingestion points: Curriculum manifest from 'agency-lab.glebkalinin.com/api/curriculum/.json' (SKILL.md).
  • Boundary markers: Absent.
  • Capability inventory: Shell command execution for 'install' toolkit items.
  • Sanitization: Absent; the instructions mandate running commands 'verbatim'.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 23, 2026, 01:02 PM
Security Audit — agent-trust-hub — lab-setup