dev-browser

Browser automation via MCP tools. ALWAYS use these tools for ANY web task - navigating sites, clicking, typing, filling forms, taking screenshots, or extracting data. This is the ONLY way to control the browser.

Quality

80%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Critical

Do not install without reviewing

Fix and improve this skill with Tessl

tessl review fix ./packages/agent-core/mcp-tools/dev-browser/SKILL.md

Quality

Content

77%

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

The body is highly actionable with strong, verification-driven workflow guidance, but it is verbose from repeated examples and fails to use its own bundle files, leaving a large monolithic document where progressive disclosure is underexploited.

Suggestions

Deduplicate repeated content — the login flow appears in the browser_script, Workflow, and Examples sections; keep one canonical example and reference it elsewhere.

Link the provided bundle files from the body (e.g., point the data-extraction/batch discussion to references/scraping.md and document the scripts/ entry points) and move the per-tool reference and error tables into a separate reference file so SKILL.md stays a lean overview.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient operational guidance, but it is padded with duplication: the complete login flow appears three times and the browser_script action list is restated in the Tools section, so it could be tightened considerably.	2 / 3
Actionability	Guidance is concrete and copy-paste ready: full tool signatures with parameters, JSON browser_script action arrays with real selectors, and an error-recovery table mapping each error to a specific remediation.	3 / 3
Workflow Clarity	Multi-step workflows are explicitly sequenced with validation checkpoints and feedback loops ('After EVERY action, verify it succeeded', the numbered verification/error-recovery flows, and the post-click tab checklist), matching the highest anchor.	3 / 3
Progressive Disclosure	Sections are clearly headed and organized, but the ~420-line SKILL.md is monolithic and none of the provided bundle files (references/scraping.md, scripts/start-relay.ts, scripts/start-server.ts) are referenced or linked from the body, so relevant reference material is never signaled.	2 / 3
	Total	10 / 12 Passed

Description

82%

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

The description clearly states capability and trigger guidance with natural action terms, but weakens itself with second-person imperative voice and maximalist over-claims ('ANY web task', 'ONLY way') that inflate conflict risk and read as fluff.

Suggestions

Rewrite in third person (e.g., 'Automates the browser via MCP tools for navigating sites, filling forms, taking screenshots, and extracting data') and drop the imperative 'ALWAYS use these tools' to avoid the voice penalty.

Replace the over-claim 'for ANY web task' / 'ONLY way to control the browser' with specific user-mentionable triggers (e.g., 'Use when the user asks to navigate a site, fill out a web form, scrape a page, or take a browser screenshot') to sharpen distinctiveness and reduce conflict risk.

Dimension	Reasoning	Score
Specificity	It lists multiple concrete actions ('navigating sites, clicking, typing, filling forms, taking screenshots, or extracting data') which alone would score 3, but the second-person imperative 'ALWAYS use these tools for ANY web task' triggers the rubric's voice penalty, reducing specificity by one.	2 / 3
Completeness	It answers both 'what' (browser automation via the listed actions) and 'when' with explicit trigger guidance ('ALWAYS use these tools for ANY web task', 'This is the ONLY way to control the browser'), so the missing-trigger cap at 2 does not apply.	3 / 3
Trigger Term Quality	Natural user-facing action terms are well covered ('navigating sites, clicking, typing, filling forms, taking screenshots, or extracting data', 'browser', 'web task'); these are phrases a user would naturally say when needing browser automation, despite 'MCP tools' being jargon.	3 / 3
Distinctiveness Conflict Risk	The 'Browser automation via MCP tools' niche with concrete actions is somewhat distinct, but the over-claims 'ANY web task' and 'ONLY way to control the browser' make it overly broad and prone to overlapping with other web-related skills.	2 / 3
	Total	10 / 12 Passed

Validation

100%

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 — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: accomplish-ai/coworker
Path: packages/agent-core/mcp-tools/dev-browser/SKILL.md
Commit: 2cf74d0

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.