The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates shell command execution to perform YouTube searches and interact with the NotebookLM CLI. Specifically, Step 2 executes a Python script using user-supplied topic keywords, and Step 5b uses the notebooklm command with dynamically generated idea angles. This pattern presents a risk of command injection if the agent does not correctly sanitize user inputs before they are incorporated into the shell commands.
[PROMPT_INJECTION]: The workflow involves ingesting and processing data from external, untrusted sources including YouTube search results and research files, which makes the skill vulnerable to indirect prompt injection.
Ingestion points: Research files in the research/ and projects/ directories, YouTube search results via a helper script, and output from the NotebookLM CLI.
Boundary markers: There are no explicit markers or instructions provided to the agent to help distinguish between trusted skill instructions and potentially malicious content within the ingested data.
Capability inventory: The skill has the capability to execute shell commands (python, notebooklm) and write files to the local vault.
Sanitization: The skill does not define any sanitization, filtering, or validation steps for external content before it is interpolated into prompts or used in logic.

ideation