The Agent Skills Directory

[SAFE]: The skill outlines a standard developer workflow for benchmarking project components. It instructs the agent to perform routine tasks such as reading repository files (.github/workflows/ci.yml, journeys_*.go) and executing local test commands (go test, gentle-ai-bench).- [COMMAND_EXECUTION]: The skill references the execution of go test and a local binary gentle-ai-bench. These operations are contextually appropriate for a benchmarking skill and are directed at the local build environment rather than remote or unauthorized systems.- [INDIRECT_PROMPT_INJECTION]: The skill defines a process where the agent reads and acts upon information in repository files. While these files represent an external data ingestion surface, this is an expected part of the agent's role in a software project, and no specific malicious injection patterns were observed in the skill's instructions.

gentle-ai-bench