The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/headless_draft.py dynamically loads and executes Python components from a separately installed engine using runpy.run_path. It includes a security pinning mechanism that validates the SHA-256 hash of every loaded file against a hardcoded manifest, preventing the execution of modified or unauthorized core code.
[DYNAMIC_EXECUTION]: The ASR (Speech-to-Text) wrapper in scripts/asr_once.py dynamically loads a local transcription script via runpy.run_path. This target script is determined by the YICHEN_ASR_EXECUTOR environment variable or a default path in the user's home directory.
[COMMAND_EXECUTION]: The skill utilizes the subprocess module to execute media utilities such as ffmpeg and ffprobe for analyzing video and rendering audio reviews. It also uses subprocess to trigger the local ASR executor script. All commands use argument lists rather than shell strings to mitigate command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests complex structured data via edit-plan.json files which determine video cutting logic.
Ingestion points: scripts/edit_plan.py reads and parses user-supplied JSON plan files.
Boundary markers: The skill requires explicit confirmation flags (settings_confirmed, check_decisions_pending) within the plan and instructs the agent to centrally resolve unclear semantic decisions with the user.
Capability inventory: The processing scripts have the capability to write to local work directories and execute media processing commands via ffmpeg.
Sanitization: The compiler script performs extensive validation on input data, including strict checks for numeric types, finite values, and valid frame-aligned time ranges.

yichen-jianying-edit