The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses curl to fetch repository tarballs from the GitHub API (api.github.com) using a provided $GH_TOKEN for authentication. While this is a network operation, it targets a well-known service (GitHub) as part of its core functionality.
[COMMAND_EXECUTION]: The skill executes multiple shell commands for environment setup and repository processing, including uv venv, uv pip install, tar -xzf, and running local Python scripts (treesit.py, gather.py, bm25.py). These are standard operations for a codebase analysis tool.
[PRIVILEGE_ESCALATION]: The skill uses chmod +x internally (referenced in metadata logic) and manages a virtual environment (uv venv). While these modify file permissions and environment paths, they are scoped to the agent's work directory and are standard for script execution.
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to ingest and analyze external, untrusted codebases. This creates a surface for indirect prompt injection if the analyzed files contain malicious instructions designed to influence the agent's reasoning during Step 4 (Feature synthesis). However, the skill uses structured analysis tools (tree-sitter) to mitigate direct execution risks, and the severity is low given the intended analytical use case.
[DYNAMIC_CONTEXT_INJECTION]: The skill documents the workflow but does not contain active !command injections in the provided files. Commands are presented as instructions for the agent to follow.

exploring-codebases