agent-browser

Drive a real browser to inspect or interact with a web page or app — navigate, take screenshots, read console and network, fill simple forms — for verification tasks, not unattended automation.

Quality

88%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

92%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

Excellent instruction-only skill content: lean, well-sectioned, concrete, and sequenced with validation checkpoints and failure-recovery guidance. The only meaningful gap is the absence of any concrete launch mechanism — no library name, command, or snippet for actually starting the controlled browser.

Suggestions

Name the automation tool in the body (puppeteer/playwright) and add a minimal launch snippet or command — e.g., how to set the user agent and viewport in one concrete call.

Add a one-line fallback for the most common wait patterns (e.g., the exact wait-on-selector / network-idle API calls) so the session steps are fully executable rather than procedural.

Dimension	Reasoning	Score
Conciseness	The body is lean with no padding and no explanations of concepts Claude already knows — every line adds skill-specific judgment (wait strategies, evidence formats, politeness limits, failure modes). It fits the "every token earns its place" anchor.	5 / 5
Actionability	Concrete parameters appear throughout ("1366×768 desktop, 390×844", "30s timeout", "retry once with a longer timeout, then escalate") plus a numbered session workflow. It stops short of 5 because the body never names or shows how to launch the automation tool — puppeteer/playwright appear only in frontmatter tags, leaving a minor execution gap.	4 / 5
Workflow Clarity	"Driving the browser" gives a clearly sequenced 7-step workflow with explicit validation (wait for a specific selector or network-idle before capturing), evidence checkpoints, and a failure-modes section with error-recovery guidance. This is a read-mostly verification skill, so the destructive/batch cap does not apply.	5 / 5
Progressive Disclosure	A single compact file with well-organized sections and no external bundle files; nothing in it is bulky enough to warrant splitting into references, so the simple-skill organization standard is met despite being slightly over 50 lines.	5 / 5
	Total	19 / 20 Passed

Description

83%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong description: concrete, third-person, and clearly niche-positioned with an explicit out-of-scope boundary. Its main weakness is the trigger side — the when is expressed as a purpose clause rather than explicit "Use when…" trigger phrases, and some natural synonyms are absent.

Suggestions

Add an explicit trigger clause, e.g., "Use when you need a screenshot of a deployed page, a UI bug repro with console/network evidence, or content that curl cannot see."

Include one or two natural synonyms users would say (e.g., "verify a UI change", "JS-rendered/SPA content") to improve trigger matching.

Dimension	Reasoning	Score
Specificity	"navigate, take screenshots, read console and network, fill simple forms" lists four distinct concrete actions spanning inspection and interaction, matching the comprehensive-coverage anchor; the actions cover the skill's full stated scope, so the minor-gaps anchor (4) does not fit.	5 / 5
Completeness	The what is explicit and concrete, and a when is present as "for verification tasks, not unattended automation" — a stated purpose boundary rather than an implied one, so not 3. It lacks an explicit "Use when…" trigger clause with concrete trigger phrases, so not 5.	4 / 5
Trigger Term Quality	Natural user phrases like "browser", "take screenshots", "web page or app", "console", "network", and "verification" are present, but common synonyms a user might say ("scrape", "SPA", "click", or the library names "puppeteer"/"playwright" that only appear in tags) are missing.	4 / 5
Distinctiveness Conflict Risk	"Drive a real browser… for verification tasks, not unattended automation" carves a clear niche (supervised browser verification) and explicitly excludes scraping, minimizing overlap with scraper or static-fetch skills.	5 / 5
	Total	18 / 20 Passed

Validation

93%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: paperclipai/paperclip
Path: packages/skills-catalog/catalog/optional/browser/agent-browser/SKILL.md
Commit: 1c07b59

Reviewed: 1 day ago

Table of Contents

Discovery Implementation Validation

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.