The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple external sources which could potentially contain malicious instructions targeting the agent during report generation.
Ingestion points: The skill ingests data from user-uploaded CSV files (references/common-csv-parsing.md), third-party integrations such as Stripe or Linear, and direct REST API calls (references/common-data-sources.md).
Boundary markers: The instructions do not specify the use of delimiters or specific safety instructions when interpolating retrieved data into the narrative analysis reports generated in Step 6 (references/detailed-analysis.md).
Capability inventory: The skill possesses significant capabilities including artifact creation (createArtifact), package installation (pnpm install), file system writes (writing React components and Express routes), and the ability to launch asynchronous design subagents (startAsyncSubagent).
Sanitization: The skill explicitly includes a security recommendation in references/common-csv-parsing.md to validate or sanitize filenames before constructing paths for server-side parsing to prevent directory escape.
[EXTERNAL_DOWNLOADS]: The skill installs several third-party libraries via pnpm during the bootstrap phase.
Evidence: The skill installs react-csv, papaparse, @tanstack/react-table, and d3-scale-chromatic from the standard npm registry (references/common-bootstrap.md). These are well-known, established packages for the intended functionality.
[DYNAMIC_EXECUTION]: The skill utilizes automated code generation and subagent workflows to build the application.
Evidence: The workflow involves running OpenAPI codegen (pnpm --filter @workspace/api-spec run codegen) and launching a specialized design subagent (startAsyncSubagent) to implement the frontend components based on the planned API surface (references/common-bootstrap.md).

data-visualization