The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The package.json and package-lock.json files contain non-existent version numbers for multiple standard Node.js dependencies, including zod@4.4.3, typescript@7.0.2, commander@15.0.0, express@5.2.1, and @modelcontextprotocol/sdk@1.30.0. Since these versions are not available on the public NPM registry, this indicates a high risk of dependency confusion where a malicious package could be served from a private or misconfigured registry.
[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes external API documentation from vendor URLs, providing a surface for indirect prompt injection if those sources are compromised.
Ingestion points: External Markdown content is fetched in src/core/tools/docs-search-tool.ts via the docs_search tool.
Boundary markers: Absent; the external content is provided to the agent's context without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has network access via api_request and the agent is instructed in SKILL.md to execute shell commands (npx).
Sanitization: Documentation is stripped of code examples via stripCodeExamples, but the natural language text is not filtered or sanitized for malicious instructions.

dataforseo-mcp-server