web-search-and-fetch
Run Google searches and fetch JS-rendered web pages as Markdown via `google-search-cli`, a Patchright-based CLI invoked through `uvx` from its GitHub repo (no local install needed). Use when the agent needs (1) fresh Google search results from a query, (2) the Markdown of a URL that plain HTTP fetch (curl/WebFetch) cannot render because it requires JavaScript or evades bots, or (3) inspection of the raw HTML of a Google results page. Trigger phrases include "google for ...", "search the web for ...", "fetch this page as markdown", "this page needs JS to render". Do not use for static doc URLs that WebFetch handles cleanly, or when Chromium cannot be installed on the machine.
72
88%
Does it follow best practices?
Impact
—
No eval scenarios have been run
Critical
Do not install without reviewing
Quality
Discovery
100%Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.
This is an excellent skill description that clearly articulates specific capabilities, provides rich natural trigger terms, explicitly states both when to use and when not to use the skill, and carefully distinguishes itself from overlapping tools like WebFetch/curl. The description is comprehensive yet focused, with numbered use cases and explicit trigger phrases that make skill selection straightforward.
| Dimension | Reasoning | Score |
|---|---|---|
Specificity | Lists multiple specific concrete actions: running Google searches, fetching JS-rendered web pages as Markdown, inspecting raw HTML of Google results pages. Also names the specific tool (`google-search-cli`), technology (Patchright-based CLI), and invocation method (`uvx`). | 3 / 3 |
Completeness | Clearly answers both 'what' (run Google searches, fetch JS-rendered pages as Markdown, inspect raw HTML) and 'when' with explicit numbered trigger conditions and trigger phrases. Also includes explicit 'do not use' guidance, which further clarifies when to select this skill. | 3 / 3 |
Trigger Term Quality | Excellent coverage of natural trigger terms: 'google for ...', 'search the web for ...', 'fetch this page as markdown', 'this page needs JS to render', plus technical terms like 'JavaScript', 'Google search results', 'Markdown', and contrast terms like 'curl/WebFetch'. These closely match what users would naturally say. | 3 / 3 |
Distinctiveness Conflict Risk | Highly distinctive: explicitly differentiates itself from plain HTTP fetch tools (curl/WebFetch) by specifying its niche is JS-rendered pages and Google searches. The 'Do not use for static doc URLs that WebFetch handles cleanly' clause directly reduces conflict risk with other web-fetching skills. | 3 / 3 |
Total | 12 / 12 Passed |
Implementation
77%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This is a well-crafted skill with excellent actionability — the exact command templates, output contracts, and error handling paths give Claude everything needed to use the tool correctly. The workflow and failure policy sections are particularly strong with explicit validation checkpoints and retry constraints. The main weakness is moderate verbosity: some sections explain internal design rationale (asymmetric headless defaults, why the install command uses the same --from URL) that Claude doesn't need to understand to use the tool effectively.
Suggestions
Trim the 'Critical default quirks' section to just state the defaults and when to override, removing the rationale about anti-bot evasion and rendering behavior.
Remove explanatory asides like 'Routing the install through the same --from URL ensures the patchright version...' — Claude doesn't need to understand why, just what to do.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Generally efficient but includes some unnecessary explanations (e.g., explaining what patchright is, the asymmetric headless defaults rationale, and the Chromium cache location). The 'Critical default quirks' section explains internal design decisions Claude doesn't need. However, most content is functional and relevant. | 2 / 3 |
Actionability | Provides exact, copy-paste-ready bash commands with all flags documented. The command templates are fully executable, output formats are specified precisely (JSON schema, Markdown with truncation marker), and examples map natural language requests to concrete commands. | 3 / 3 |
Workflow Clarity | The workflow section provides a clear numbered sequence with explicit validation/error handling: checking for truncation, handling anomalous output with retry (with a 'do not loop' constraint), and distinct error paths for exit code ≠ 0. The failure policy section adds additional guardrails including rate-limit detection and a clear 'stop and report' policy. | 3 / 3 |
Progressive Disclosure | Content is well-organized with clear section headers (Prerequisites, Commands, Output contract, Workflow, Failure policy, Examples), but everything is in a single file with no references to external documentation. Some sections like 'Critical default quirks' and 'Browser-state cache' could be separated or condensed, though the skill is borderline in length where a single file is acceptable. | 2 / 3 |
Total | 10 / 12 Passed |
Validation
100%Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.
Validation — 11 / 11 Passed
Validation for skill structure
No warnings or errors.
- Repository
- ceshine/ceshine-agent-skills
- Commit
fc251b2
- Reviewed
Table of Contents
Is this your skill?
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.