The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with a local command-line interface tool named jianying. It executes queries such as jianying capabilities --json and jianying doctor --json to assess the local environment and ensure the platform supports required editing features.
[EXTERNAL_DOWNLOADS]: The instructions guide the user or agent to install supplementary capabilities using the npx skills add command. These resources (e.g., jianying-draft-schema, jianying-draft-edit) belong to the same vendor (full-aigc-skills) as the skill author, representing standard modular installation within a trusted ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data, such as media metadata and project goals, to select scenarios and recipes. While this creates a vulnerability surface for indirect injection, the skill implements multiple safety guardrails, including explicit instructions to avoid fictionalizing data and requiring manual user approval at critical decision points.
Ingestion points: User-provided media facts, channel requirements, and audience information entering the context through SKILL.md and scenario templates.
Boundary markers: The skill uses structural JSON inputs for job handovers but lacks high-entropy delimiters (like XML tags) for raw user input.
Capability inventory: The skill generates project schemas and timeline plans which are then executed by external CLI tools or sibling skills.
Sanitization: Relies on human-in-the-loop verification checklists and automated capability handshakes before any persistent changes are made.

jianying-draft-video-planning