The Agent Skills Directory

[PROMPT_INJECTION] (HIGH): The skill provides patterns for MCP server tool handlers that are vulnerable to Indirect Prompt Injection (Category 8).
Ingestion points: Untrusted data enters the agent context through the input parameter in the searchTool example in SKILL.md.
Boundary markers: No boundary markers or 'ignore embedded instructions' warnings are present in the provided tool handler templates.
Capability inventory: The tool demonstrates a database write/execute capability via db.query(args.query), which maps to runQuery(sql, config) in the provided service layer.
Sanitization: While structural validation is performed via @effect/schema, the resulting data is interpolated directly into a query string, creating a significant SQL injection surface.
[EXTERNAL_DOWNLOADS] (MEDIUM): The skill requires the installation of multiple external packages (effect, @effect/schema, @effect/opentelemetry) from the effect-ts organization. While widely used, this organization is not on the provided Trusted External Sources list, making the dependencies unverifiable by strict policy.
[MALICIOUS_URL_ALERT] (INFO): An automated scan flagged Layer.su as a phishing URL. This is identified as a false positive; the string is a prefix of the common library method Layer.succeed used in the code snippets, and no such domain is actually referenced in the file content.

effect-ts