The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow instructs the orchestrator agent to use standard git commands (git ls-remote, git clone, git remote get-url) to manage external source code sources.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading repository content into a local .slim/clonedeps/repos/ directory. The instructions include validation logic to ensure URLs are HTTPS-based and do not contain embedded credentials or local path references.
[SAFE]: The skill implements several defensive measures to mitigate risks associated with processing external content:
Script Execution Prohibition: Explicitly instructs the agent not to run installation, build, or test scripts from cloned repositories.
Path Sanitization: Employs a safe-naming convention for directories (e.g., replacing slashes with double underscores) to prevent path traversal.
User Control: Requires the agent to present a plan and obtain user confirmation before performing network cloning.
Contextual Isolation: Adds clear marker blocks to AGENTS.md and .ignore files to identify cloned content as read-only metadata, which helps mitigate indirect prompt injection surfaces by providing clear boundaries for future agent interactions.

clonedeps