The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Node.js commands (node packages/skill-tools/dist/index.mjs) to perform grading and benchmarking. These operations target scripts within the local project structure and are part of the intended workflow for a testing utility.
[EXTERNAL_DOWNLOADS]: Contains a documentation reference to a GitHub repository (github.com/zwbao/skill-creator-pro). This is used for attribution and technical reference, posing no security risk.
[PROMPT_INJECTION]: The skill implements an evaluation loop that interpolates external data (evals.json) into prompts for a subagent. While this represents an indirect injection surface, it is the primary purpose of the skill (testing and evaluation) and is contained within the development workspace environment.
[DATA_EXFILTRATION]: The workflow involves reading skill configurations and writing execution transcripts to a .workspace/ directory. All file operations are local to the repository, and no network exfiltration patterns were detected.

skill-eval