effectcraft-harness

Pass

Audited by Gen Agent Trust Hub on Oct 13, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads required software artifacts from GitHub and Python.org during its initialization phase.
  • Fetches the EffectCraft CLI binary from the vendor's official GitHub repository (storytold/effectcraft).
  • Fetches standalone Python distributions from well-known sources, including astral-sh/python-build-standalone and the official Python.org FTP server.
  • [REMOTE_CODE_EXECUTION]: The harness executes downloaded binaries and scripts on the host system to perform media automation and processing.
  • To mitigate risks, every downloaded artifact is verified using a SHA256 checksum against hardcoded values in local lockfiles (runtime.lock.json, python.lock.json) before execution.
  • [COMMAND_EXECUTION]: Extensively utilizes subprocess.run and subprocess.Popen to manage external tool lifecycles and conduct diagnostic probes.
  • These operations are used for process isolation, allowing the harness to monitor and terminate worker processes independently of the main agent session.
  • [DYNAMIC_EXECUTION]: Employs dynamic JavaScript generation to communicate with the application's media engine via the Model Context Protocol (MCP).
  • Snippets are constructed locally to query read-only information, such as project revision history or current UI state, and do not incorporate unvalidated external input into executable code.
  • [INDIRECT_PROMPT_INJECTION]: The harness processes external plans and user-provided inputs, creating a potential attack surface.
  • Ingestion points: Ingests structured JSON plans and paths to input media files via the managed.py and workflow.py scripts.
  • Boundary markers: The skill enforces strict data separation by using structured JSON objects for command parameters.
  • Capability inventory: The skill possesses broad capabilities for filesystem interaction and application control within the EffectCraft environment.
  • Sanitization: Implements rigid JSON schema validation for all tool parameters and verifies the integrity of input assets using SHA256 digests before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 13, 2026, 01:30 PM