effectcraft-cli-masks
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads and installs the EffectCraft CLI, AI models for roto-brushing and face tracking, and a desktop application from external repositories (e.g., GitHub). This process is mitigated by strict verification of SHA-256 checksums, binary signatures, and file sizes defined in the skill's metadata (references/runtime.lock.json, references/bridge-tools.json).
- [DYNAMIC_EXECUTION]: The skill includes a 'run_script' tool and 'file.runScript' command that allow for the execution of JavaScript (After Effects-style scripting) within the engine environment. This is a core capability for complex video editing tasks but represents a dynamic code execution surface.
- [COMMAND_EXECUTION]: The 'execute_command' tool provides the ability to run arbitrary internal engine commands. Access is managed through a managed launcher that enforces isolation and logs all operations in a journal for auditing.
- [INDIRECT_PROMPT_INJECTION]: The skill processes complex JSON-based 'command plans' and 'workflow templates' (e.g., examples/brand-intro.json). This constitutes an attack surface if untrusted data is incorporated into these plans. However, the skill implements structural validation (e.g., duplicate key checks) and a visual 'Judge' review process (references/codex-judge.md) to verify outcomes before acceptance.
- [REMOTE_CODE_EXECUTION]: The 'file.executeFile' command provides a mechanism to execute files. While potentially high-risk, it is presented as a tool capability with an explicit 'confirmed' parameter, suggesting user oversight is required for its use.
Audit Metadata