The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/collect_sglang_review_corpus.py communicates with GitHub's official API (api.github.com) to collect PR review data. This is a well-known technology service used for its intended purpose.
[COMMAND_EXECUTION]: The collection script uses subprocess.run to call the gh CLI for retrieving an authentication token (gh auth token). This is a standard and safe practice for developers to authenticate with GitHub services.
[INDIRECT_PROMPT_INJECTION]: The skill processes historical GitHub PR comments which could theoretically contain malicious instructions. However, the risk is categorized as low because the scripts strictly perform search and summarization tasks, and the agent's workflow treats the data as reference evidence for code review rather than executable instructions.
[CREDENTIALS_UNSAFE]: The scripts safely handle credentials by reading from environment variables or standard CLI authentication tools (GITHUB_TOKEN, gh auth token), adhering to secure configuration practices.

sglang-humanize-review