The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes git and gh (GitHub CLI) commands to initialize repositories, clone sources, and synchronize changes with remotes. These operations are restricted to the ~/.gtm/ path and are governed by explicit user approval steps defined in the workflow instructions.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests content from external URLs and local files to generate workspace documentation.
Ingestion points: Processes user-provided links, social media profiles, and local file paths during the workspace intake and update procedures.
Boundary markers: The skill requires an 'accept loop' for every durable change, where the agent must present the complete proposed content and file operations for user review and approval before writing to disk.
Capability inventory: Includes file system writing within the dedicated workspace directory and the ability to execute repository-related shell commands.
Sanitization: A 'Link safety' policy (detailed in references/contract.md) explicitly instructs the agent to detect and refuse URLs containing credentials, tokens, or session identifiers to prevent accidental exposure or persistence of secrets.

gtm-workspace