The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads human-readable workflow documentation from an external file to drive code generation, which could be manipulated to include malicious instructions.
Ingestion points: The skill uses Glob and Read to ingest data from /workflows/mobile-workflows.md.
Boundary markers: It uses structural parsing based on markdown headings and specific HTML-style comments (e.g., ), but does not provide explicit isolation or instruction-filtering for the content within the workflow steps.
Capability inventory: The skill has the ability to write to the local filesystem (e2e/mobile/), spawn a subagent (Explore agent) for selector discovery, and execute shell commands via npx for type-checking.
Sanitization: The skill lacks explicit sanitization or escaping of the workflow step strings before they are translated into executable Playwright test code.
[DYNAMIC_EXECUTION]: The skill dynamically generates a complete Playwright project and then executes validation tools on the generated source.
It generates multiple configuration and test files (playwright.config.ts, auth.setup.ts, workflows.spec.ts) based on external inputs.
It performs runtime validation by executing npx tsc --noEmit to type-check the generated code.
It includes a self-correcting logic loop that parses compiler errors and modifies the generated files on disk to resolve issues.
[COMMAND_EXECUTION]: The skill executes shell commands to set up the environment and validate the generated project.
It runs npm install, npx playwright install, and npx tsc within the context of the generated e2e/mobile/ directory.

mobile-workflow-to-playwright