The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted transcript and video data, which represents a potential attack surface for indirect prompt injection.
Ingestion points: Transcript files (assets/transcript.json), source video files, and reference reels are read by the agent to shape the editorial direction and timing.
Boundary markers: The instructions do not specify explicit delimiters or "ignore embedded instructions" guards for the transcript processing phase.
Capability inventory: The skill uses local scripts (node, powershell) for validation, writes project artifacts to the file system, and performs network operations via external APIs.
Sanitization: No specific sanitization or filtering of transcript content is mentioned before it is used to influence the agent's editorial decisions.
[EXTERNAL_DOWNLOADS]: The skill interacts with external services such as ElevenLabs for transcription and Kie.ai for B-roll generation.
Evidence: Instructions in SKILL.md direct the agent to use these services and refer to docs/TOOLS-AND-API-KEYS.md for credentials.
Context: These interactions are central to the skill's primary function and follow the author's specified tool chain.
[COMMAND_EXECUTION]: The skill uses local Node.js scripts to validate project plans and footage assets.
Evidence: SKILL.md includes commands to run scripts/validate-plan.mjs and scripts/validate-footage.mjs using node.
Context: These scripts are internal to the skill and perform safety/quality checks on project metadata and file hashes.

short-form-edit