The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The script scripts/collect_sglang_review_corpus.py accesses the GITHUB_TOKEN environment variable and executes the GitHub CLI (gh auth token) to facilitate authenticated API requests to api.github.com. These operations are used to fetch public pull request data for the SGLang project. As this targets a well-known service and aligns with the skill's primary function, it is considered safe practice for developer tooling.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external code diffs and historical PR comments. Although the workflow lacks explicit prompt boundaries, the use of helper scripts to digest the corpus reduces the risk of the agent directly executing instructions embedded in the analyzed text. This represents a standard attack surface for code analysis tools and is handled safely within the intended workflow.

sglang-humanize-review