The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestrator script scripts/bianpai.mjs utilizes child_process.spawnSync to execute various system commands and local scripts.
Evidence: Spawns node, bash, ffprobe, and ffmpeg to perform metadata extraction, contract verification, and vendor synchronization.
Mitigation: Command paths are constructed relative to the project root or via environment variables (LAOHAN_SKILLS_ROOT), limiting execution to known local scripts.
[DYNAMIC_EXECUTION]: The skill dynamically executes sub-validators and specialized tools within the production ecosystem at runtime.
Evidence: Executes scripts such as check-script-contract.mjs and check-topic-contract.mjs located in peer skill directories.
Context: As an orchestrator, this behavior is central to its function. The execution is restricted to the vendor's own suite of tools (hanzhcn).
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as its orchestration logic depends on the contents of various project files that could be influenced by external data.
Ingestion points: Reads numerous JSON and Markdown files including 00-选题.json, 01-口播稿.md, and 09-导演/director-state.md to determine the production state.
Boundary markers: The skill employs strict "mechanical contracts" and state markers (e.g., CLEAR, COMPLETED, PARTIAL) to delimit valid data.
Capability inventory: The script can spawn subprocesses (Node.js/Bash) based on the state determined from ingested files.
Sanitization: Implements high-quality sanitization by using realpathSync and prefix-matching to prevent path traversal when reading files (e.g., realpathSync(target).startsWith(episodeRealRoot + '/')). It also enforces strict SHA-256 integrity checks for production artifacts.

laohan-bianpai