The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/evaluation.py and scripts/connections.py files allow for the execution of local shell commands via the stdio transport. This is a primary intended feature of the skill, enabling developers to launch and test their MCP server implementations locally.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external MCP servers and XML evaluation files, creating a potential surface for indirect prompt injection. A malicious MCP server could return content designed to influence the agent's behavior during the evaluation process.
Ingestion points: The evaluation.py script reads from the user-provided eval_file and receives results from tool calls to the MCP server under test.
Boundary markers: The system prompt in evaluation.py instructs the agent to wrap its output in <summary>, <feedback>, and <response> tags.
Capability inventory: The script can execute local commands, perform network operations via SSE/HTTP, and call the Anthropic API.
Sanitization: External data from tool calls and XML files is included in the prompt context without explicit sanitization or filtering.
[EXTERNAL_DOWNLOADS]: The skill's documentation suggests fetching official specifications and SDK documentation from official Model Context Protocol domains and GitHub repositories. These are well-known and trusted sources used for reference and documentation purposes.

mcp-builder