The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python diagnostic script (diagnose.py) to verify the status of built-in search tools. It also instructs the agent to use various shell commands and tools like grep, find, and awk to inspect and modify configuration files.
[EXTERNAL_DOWNLOADS]: The skill configures connections to external Model Context Protocol (MCP) servers such as Exa and Tavily. These are well-known search services. It also references trusted cloud platforms like Google Cloud Vertex AI and Amazon Bedrock.
[REMOTE_CODE_EXECUTION]: The skill recommends the use of npx one-search-mcp, which involves downloading and executing a package directly from the npm registry to provide local search gateway functionality.
[CREDENTIALS_UNSAFE]: The diagnostic script and instructions involve reading sensitive information, including API keys and authentication tokens, from local configuration files such as ~/.claude/settings.json and ~/.codex/auth.json. This access is required to probe the functionality of the configured model endpoints.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from external, untrusted web search results into the agent's context.
Ingestion points: Data is ingested through replacement MCP search tools (e.g., Exa, Tavily, Bocha) described in SKILL.md and backends.md.
Boundary markers: The instructions do not specify explicit boundary markers for the data returned by these external providers.
Capability inventory: The skill has the capability to modify agent permissions (permissions.allow/deny), write to local configuration files, and execute shell commands.
Sanitization: No specific sanitization or filtering protocols for external content are defined within the skill instructions.

agent-web-search-setup