The Agent Skills Directory

[PERSISTENCE]: The skill provides instructions in references/setup.md to modify the user's shell profile (~/.zshrc) to export the TUDUDI_API_TOKEN environment variable. Modifying shell profiles to execute code or set variables on startup is a documented persistence mechanism.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted meeting transcripts, which creates a vulnerability surface where malicious instructions could be embedded in the transcript data.
Ingestion points: External session transcripts retrieved via the anarlog CLI tool, as described in references/anarlog.md.
Boundary markers: None identified; the skill does not specify delimiters or instructions for the agent to ignore potentially malicious content within transcripts.
Capability inventory: The skill can perform network writes to the Tududi API (references/api.md) and create or modify files in the docs/anarlog/ directory (references/anarlog.md).
Sanitization: None identified; there is no mention of validating, filtering, or escaping the transcript content before it is processed or written to other locations.
[COMMAND_EXECUTION]: The skill executes local system commands using the anarlog CLI to list, retrieve, and archive meeting data. It also references the use of security find-generic-password on macOS to retrieve secrets from the system Keychain.
[EXTERNAL_DOWNLOADS]: The skill interacts with an external API at a configurable base_url to synchronize repository state and tasks. It uses bearer token authentication to perform these network operations.

tududi