The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads content from external sources as part of its research pipeline. It utilizes git clone in scripts/clone_repo.py to download repositories from GitHub for analysis. It also fetches metadata and README files via the GitHub API using the gh CLI in scripts like repo_metadata.py and repo_readme_fetch.py. Additionally, it queries the Papers With Code API in scripts/search_paperswithcode.py using urllib.request. These targets are all well-known and trusted services.
[COMMAND_EXECUTION]: The skill executes system commands to interact with GitHub and the local file system. It spawns git and gh processes via subprocess.run in scripts/clone_repo.py and scripts/repo_metadata.py. These calls use argument lists rather than shell strings, which is a security best practice that prevents command injection.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it is designed to ingest and summarize untrusted data from external GitHub repositories and academic databases. This risk is inherent to the primary purpose of the skill (deep research) and is mitigated by the underlying agent's instruction-following guardrails.

github-research