openmotion-recreate
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided content from the workspace and reference videos to generate film projects.\n
- Ingestion Points: Instructions in
SKILL.mddirect the agent to read workspace files (README, docs, marketing copy) and external reference videos to determine film content.\n - Capability Inventory: The agent can edit files via
openmotion_edit_project, execute media processing commands via shell subprocesses, and request project exports.\n - Boundary Markers: No explicit delimiters or specific instructions to disregard embedded commands are present when reading external workspace data.\n
- Sanitization: There is no evidence of sanitization or validation of the ingested text or video metadata to prevent injection attacks.\n- [DYNAMIC_EXECUTION]: The skill utilizes dynamic loading and script wrapping techniques to manage its execution environment.\n
- Evidence:
scripts/vision-tools.mjsuses a dynamicimport()call to load a runtime configuration script (vision-runtime.mjs) from a path calculated at execution time.\n - Evidence: It employs a Python wrapper using
runpy.run_pathto execute local script files includingmeasure_motion.pyandobserve_marks.py.\n- [EXTERNAL_DOWNLOADS]: The skill's setup procedure involves downloading binary assets from the vendor's infrastructure.\n - Evidence: Documentation in
reconstruction.mdandvision-tools.mjsdescribes a preparation step that downloads verified platform assets (Python runtime, NumPy, OpenCV) into a local cache to support media analysis.\n- [COMMAND_EXECUTION]: The skill executes external binaries to perform its primary technical functions.\n - Evidence:
scripts/measure_frame.mjsspawnsffmpegandffprobeprocesses to extract and analyze video frames.\n - Evidence:
scripts/vision-tools.mjsspawns apythonprocess to run analysis scripts.
Audit Metadata