The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a tool that processes user-supplied inputs and interacts with Sanity Studio data via a client. This creates a surface where malicious instructions could be embedded in data or inputs to influence the agent's behavior.
Ingestion points: The inputs object in the run() function (documented in SKILL.md) accepts string values from the UI.
Boundary markers: No specific boundary markers or delimiters are defined in the documentation to separate instructions from data.
Capability inventory: Scripts have access to the client (Sanity client) to perform data operations and a log object to output to the UI.
Sanitization: The documentation suggests manual validation ("Validate and trim values inside run() when needed") but does not provide or enforce automated sanitization.
[COMMAND_EXECUTION]: The skill provides standard development commands for verification and building within a monorepo structure.
Evidence: pnpm lint, pnpm --filter test-studio build, and pnpm --filter test-studio... build are listed as verification steps in SKILL.md.

test-studio-script-runner