The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the monologue CLI tool to search (monologue notes all) and retrieve (monologue notes get) user notes. This execution is scoped to the tool's intended use for transcript management and is triggered based on user descriptions.
[INDIRECT_PROMPT_INJECTION]: As the skill processes external transcript data, there is a potential surface for indirect prompt injection. The skill implements several safeguards against this:
Ingestion points: Transcripts provided via pasted text, file uploads, or retrieved from monologue notes (SKILL.md).
Boundary markers: The instructions use structured sections to separate processing rules from the source data, treating the source as 'evidence'.
Capability inventory: The agent can invoke the monologue CLI tool to interact with the user's notes.
Sanitization: Explicit instructions to 'preserve claims, intent, tone' and 'never fill an inaudible passage' serve as functional constraints against the agent following instructions embedded within the transcript text.
[SAFE]: The skill includes a specific security instruction to 'never request or accept an API token in chat', directing users to use the monologue onboarding command in their own terminal instead. This follows established best practices for secret management.

clean-transcript