The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses git clone to fetch source code from external repositories into a local temporary directory. This is the primary function of the skill, allowing the agent to inspect real source code when documentation is insufficient.
[COMMAND_EXECUTION]: The skill executes shell commands to manage a temporary workspace, specifically creating the .tmp/ directory and ensuring it is added to the project's .gitignore file to prevent external code from being committed to the user's repository.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it analyzes untrusted content from external repositories. The skill design explicitly mitigates this by delegating the exploration to subagents, ensuring that raw repository content does not flood the main conversation context.
Ingestion points: External source code files and repositories cloned into the .tmp/repos/ directory.
Boundary markers: The use of isolated subagents for exploration serves as a structural boundary, keeping raw content out of the main thread's context window.
Capability inventory: Filesystem modification (mkdir, .gitignore updates), network operations (git clone), and the ability to spawn subagents for delegated tasks.
Sanitization: No explicit sanitization or filtering of the content read from the cloned repositories is defined beyond the agent's inherent processing logic.

source-investigator