The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill manages the installation of the youtube-transcript-api package from the Python Package Index (PyPI). This installation is initiated in scripts/install-dependencies.sh and is only performed after explicit user confirmation in the SKILL.md workflow.
[COMMAND_EXECUTION]: Local shell commands are utilized to validate the Python environment and extract video metadata. SKILL.md instructions include the use of grep and sed for parsing URLs and executing the local extract-transcript.py script.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its handling of untrusted external video transcripts.
Ingestion points: Video transcripts are retrieved and saved to /tmp/transcript_${VIDEO_ID}.txt, which is then read directly into the agent's context in SKILL.md (Step 4).
Boundary markers: There are no explicit delimiters (such as XML tags or unique markers) or specific system instructions provided to the agent to distinguish between the transcript data and the summarization task instructions.
Capability inventory: The skill uses subprocess execution (python3, pip) and filesystem access (/tmp/ directory) to perform its core functions.
Sanitization: The transcript text is processed without sanitization or validation to detect embedded instructions before being passed to the AI model.

youtube-summarizer