skills/miguelrios/unc-skills/parable/Gen Agent Trust Hub

parable

Pass

Audited by Gen Agent Trust Hub on Jul 22, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install external tools such as the Cursor CLI via a shell script from cursor.com and research skills via npm. These sources are established technology services.
  • [COMMAND_EXECUTION]: The verify subcommand executes shell commands defined in the parable.toml configuration file to run tests and linters. It uses shlex.quote to sanitize user-provided targets, reducing the risk of command injection.
  • [COMMAND_EXECUTION]: The claude subcommand executes a local binary specified in the configuration (defaulting to 'claude') using os.execvpe.
  • [DATA_EXFILTRATION]: The usage tracking tool reads local authentication tokens for Anthropic, OpenAI, and Cursor to check subscription headroom. These tokens are transmitted only to the respective official vendor API endpoints.
  • [PROMPT_INJECTION]: The skill processes untrusted data such as git diffs and task plans. It employs a WORKSPACE_CONTRACT as a boundary marker in plans and uses shlex.quote for sanitizing shell arguments. Capability inventory includes subprocess execution for tests and binary execution for sessions.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 22, 2026, 06:55 PM
Security Audit — agent-trust-hub — parable