The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute numerous local Python scripts including smart_cut.py, transcribe_groq.py, resegment.py, draw.py, and make_vertical.py to perform video and image processing tasks.
[COMMAND_EXECUTION]: The instructions include shell commands for credential management, specifically export $(cat ~/.openai.env | xargs) to load environment variables, and directory management commands like mkdir and path cleaning.
[CREDENTIALS_UNSAFE]: The skill requires an OpenAI API key and explicitly directs the agent to access sensitive file paths such as ~/.openai.env or environment variables to retrieve it. Accessing or managing secrets in plain-text environment files on the file system carries inherent risks of credential exposure.
[PROMPT_INJECTION]: The skill processes video transcripts (ingested via transcribe_groq.py and stored as .txt) to generate marketing content, titles, and metadata. This creates an indirect prompt injection surface where instructions embedded in a video's audio could potentially influence the agent's behavior during the summarization or title generation phases.
Ingestion points: Reads transcript data from working/<video-id>/<video-id>.srt and .txt files.
Boundary markers: No explicit boundary markers or "ignore embedded instructions" warnings are used when interpolating transcript content into prompts.
Capability inventory: The skill has capabilities for file writing (output/ and working/ directories), network operations (OpenAI/Groq APIs), and subprocess execution (numerous Python scripts).
Sanitization: Basic sanitization is performed on video titles to remove Windows-incompatible characters, but no content-level sanitization is mentioned for the transcripts themselves.
[EXTERNAL_DOWNLOADS]: The skill utilizes external services for its core functionality, including the OpenAI API for image generation and the Groq API for audio transcription. These are well-known technology services.

claude-youtube-video-workflow