The Agent Skills Directory

[SAFE]: The skill operates locally within the user's workspace and specific platform configuration directories, with no external network connectivity or unauthorized file access detected.
[COMMAND_EXECUTION]: The skill leverages local shell scripts (e.g., gtd_init.sh, gtd_status.sh) to perform administrative tasks like system setup and reporting. These scripts are bundled with the skill and execute predictable, safe operations.
[INDIRECT_PROMPT_INJECTION]: The skill processes task descriptions and inbox items provided by the user, which is a standard feature for task management tools. This surface is managed by the underlying agent's safety guardrails.
Ingestion points: Reads user-supplied tasks from memory/gtd/inbox.md and processes them via prompt templates.
Boundary markers: Instructions and user data are separated by markdown headers and defined prompt blocks.
Capability inventory: Includes file system read/write, shell script execution, and interaction with system calendar/approval providers.
Sanitization: Relies on the LLM's internal safety filters and contextual processing.

gtd-harness