The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the download of an external package, @mhalder/qdrant-mcp-server@3.3.1, from the npm registry, which is a well-known service. This is used to run the core MCP server functionality.
[REMOTE_CODE_EXECUTION]: The skill instructs the agent to execute the downloaded package using npx -y. This execution pattern is necessary for the skill's primary function of providing an MCP server.
[COMMAND_EXECUTION]: The provided scripts scripts/setup-mcp.sh and scripts/check-health.sh execute various system commands, including claude mcp add, mkdir, curl, and node. These are used for configuring the agent environment and verifying infrastructure readiness.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it processes untrusted external data (the user's codebase and git history) to build a searchable index.
Ingestion points: Untrusted data enters the agent context through the index_codebase and index_git_history tools, which read source files, comments, and commit messages.
Boundary markers: The skill includes explicit warnings in SKILL.md and references/configuration.md (Security Guardrails sections) instructing the agent to treat indexed content as context only and not as trusted instructions.
Capability inventory: The skill has the capability to execute shell commands (via the setup scripts and the configured MCP server) and access the local file system (required for indexing).
Sanitization: No specific content sanitization or filtering logic is present in the provided shell scripts; the safety relies on the agent's interpretation and the warnings provided in the documentation.

qdrant-codebase-search