The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes numerous instructions for executing shell commands locally using pnpm and node. These commands facilitate building the bench environment, running specific scenarios, and managing benchmark data.
[REMOTE_CODE_EXECUTION]: The skill provides instructions to trigger GitHub Action workflows on the sanity-io/sanity repository using the gh CLI. This involves remote execution on GitHub's infrastructure, which is scoped to the author's own repository and intended for CI/CD operations.
[DYNAMIC_EXECUTION]: The skill utilizes the tsx loader to dynamically execute TypeScript scripts (perf/bench/cli/index.ts) from the local file system using node. This is a standard method for providing a CLI within the project.
[INDIRECT_PROMPT_INJECTION]: The skill outlines a surface for processing untrusted data from GitHub pull request labels and workflow dispatch inputs.
Ingestion points: The agent reads trigger:perf-bench PR labels and inputs like ab_from, ab_to, and backfill_sha to configure benchmark runs.
Boundary markers: Documentation does not mention specific delimiters or warnings for embedded instructions in the inputs.
Capability inventory: The skill has the ability to trigger GitHub workflows and execute local scripts.
Sanitization: The documentation notes that the workflow itself performs validation on the supplied inputs.

sanity-bench