agent-web-search-setup

An agent skill by daymade, from daymade/claude-code-skills. Tags: api, automation, setup, troubleshooting, web.

What it does

Sets up working web search on an agent whose model backend cannot run it. Resellers and relays that proxy Claude or Codex to another cloud (Vertex AI, Bedrock, or an OpenAI-compatible layer) do not execute the server-side web_search and web_fetch tools, so those return empty instead of failing and the model reports that recent things do not exist. Use this skill whenever searches come back with nothing, a model insists a shipped product was never released, web search appears to do nothing at all, someone wants to give an agent internet access, or the person is on a third-party base URL, a relay, or a 中转站. It works out which built-in tools are actually dead, removes them so the model stops reaching for them, installs a replacement the model can really call, and proves it with a live query. It configures the agent; it is not itself a search engine.

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add daymade/claude-code-skills --skill agent-web-search-setup

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/daymade/claude-code-skills
cp -r claude-code-skills/daymade-claude-code/agent-web-search-setup ~/.claude/skills/agent-web-search-setup

Safety box score

Grade D, elevated risk: 46 of 100 risk points, rated 2026-09-20. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill diagnoses whether web search is broken on a relay and then changes Claude/Codex configuration to disable built-in tools and register a replacement MCP backend. The main risks are live network probing with user credentials, persistent edits to home-directory config files, and changes to the agent’s allowed tool set.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	1 (low)	It ships a runnable Python probe (`scripts/diagnose.py`) and the workflow relies on shell commands like `python3 ...`, `claude mcp add ...`, and `codex exec ...`, but it does not execute arbitrary downloaded code.
Network access	2 (moderate)	`scripts/diagnose.py` POSTs to remote endpoints such as `/v1/messages` and `/v1/responses`, and the workflow explicitly performs live web queries and `curl` checks against third-party services.
Filesystem reach	2 (moderate)	It reads and writes user config in `~/.claude/settings.json`, `~/.claude.json`, `~/.codex/config.toml`, `~/.codex/auth.json`, and project `.mcp.json` / `.claude` files to register servers and change permissions.
Secrets & credentials	2 (moderate)	The probe reads API keys from `ANTHROPIC_AUTH_TOKEN`, `ANTHROPIC_API_KEY`, and `OPENAI_API_KEY`, plus config files, and sends them in request headers (`x-api-key`, `authorization`).
Privilege & persistence	1 (low)	It does not need sudo or global installs, but it does alter user-level agent settings and can move config roots/permissions in ways that persist across sessions.
Instruction hijack surface	1 (low)	The docs are highly directive and tell the agent to ignore misleading signals and use specific file sources, but there is no obvious hidden payload, remote prompt-following, or malicious instruction injection.
Irreversible actions	2 (moderate)	The skill can add MCP servers and grant `permissions.allow` entries, which persistently changes what tools the agent may call and can enable autonomous web access.
Transparency	0 (none)	All shipped files are readable markdown or Python; there are no binaries, base64 blobs, downloaded installers, or decoded-at-runtime code.

Before installing

Read `scripts/diagnose.py` and confirm the target base URL/key come from the place you expect.
Check which files it will edit: `~/.claude/settings.json`, `~/.codex/config.toml`, `~/.codex/auth.json`, and any project `.claude`/`.mcp.json` files.
Run the probe in a scratch account or with dummy credentials first, so you can see what leaves the machine.
If you use a relay, verify the configured `ANTHROPIC_*` or `OPENAI_*` values before letting it make live requests.

Rated 2026-09-20 by an automated review of 4 files from the skill's repository.

Source

Repository: daymade/claude-code-skills (all skills from this repository)
Path: daymade-claude-code/agent-web-search-setup/SKILL.md
Branch: main
Updated: 2026-09-20

chrome-mcp-troubleshooting — Diagnose and fix Claude in Chrome MCP extension connectivity issues.
gh-cli — Enforces authenticated gh CLI workflows over unauthenticated curl, WebFetch, and MCP fetch patterns.
running-release-tests — Run automated release testing (UI or API) via the AWS DevOps Agent using a pre-configured test profile.
X MCP guide — ALWAYS read this when a user connects the X plugin or any X MCP, before using any X connection, and again on any X error.
gangtise-copilot — Gangtise (岗底斯投研) OpenAPI skill suite installer and diagnostic tool.
ima-copilot — Installs, troubleshoots, and personalizes the official Tencent IMA skill (a wrapper layer that orchestrates upstream ima-skill, not a replacement).