cua-driver

Drive a native GUI app (macOS, Windows, Linux) via the cua-driver CLI (default) or MCP server; snapshot its accessibility tree, act through snapshot-bound element tokens, native menu paths, exact window geometry, or pixel coordinates, and verify from fresh state. Use when the user asks you to operate, drive, automate, or perform a GUI task in a real application on the host, or to continue, resume, or recall recent Cua activity.

Quality

88%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Quality

Content

77%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.

A thorough, highly actionable core with strong validation-gated workflows, but it is overly long with cross-section redundancy and leans on five companion files that are not bundled, leaving the progressive-disclosure structure partially non-functional.

Suggestions

Bundle the referenced MACOS.md, WINDOWS.md, LINUX.md, BROWSER.md, and RECORDING.md (or remove the references) so the one-level-deep navigation actually resolves.

Consolidate the ax/px distinction and the snapshot-before-action invariant into a single authoritative section and cross-reference it elsewhere instead of re-explaining.

Trim the Agent cursor overlay section to the operational rules (enable during control, seed/glide caveat, per-session ownership) and move detailed theming prose to a reference.

Dimension	Reasoning	Score
Conciseness	Most content is skill-specific (the ladder, tool surface) so it earns its place, but the ax/px distinction and the snapshot-before-action invariant recur across the behavior matrix, dispatch table, and 'Things to avoid', and the Agent cursor overlay section is padded with theming detail.	3 / 5
Actionability	Copy-paste-ready throughout: the canonical loop, the tool dispatch table with exact signatures, the `cua-driver mcp …` startup commands, the CLI get_window_state examples, and the PIL crosshair Python snippet are all executable and cover common cases.	5 / 5
Workflow Clarity	The verify-then-escalate ladder is an explicit sequenced procedure with validation checkpoints (`verify_state` after every action), feedback loops (validate → fix → re-validate, rung escalation), and the batch/destructive filesystem guidance requires independent readback of source and destination.	5 / 5
Progressive Disclosure	Section headers and well-signaled one-level-deep references to MACOS.md, WINDOWS.md, LINUX.md, BROWSER.md, and RECORDING.md are present, but none of those companion files exist in the bundle, so the navigation is broken and the platform-specific layer is referenced but unavailable.	3 / 5
	Total	16 / 20 Passed

Description

100%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, third-person description that explicitly covers capabilities, trigger conditions, and a distinct niche. It is comprehensive without padding and uses natural, synonym-rich trigger language.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'snapshot its accessibility tree', 'act through snapshot-bound element tokens, native menu paths, exact window geometry, or pixel coordinates', 'verify from fresh state' — matching the comprehensive-coverage anchor.	5 / 5
Completeness	Explicitly answers both what ('Drive a native GUI app … snapshot … act … verify') and when ('Use when the user asks you to operate, drive, automate, or perform a GUI task'), matching the concrete-trigger-phrases anchor.	5 / 5
Trigger Term Quality	Natural user phrases with synonym coverage — 'operate', 'drive', 'automate', 'perform a GUI task', 'continue, resume, or recall recent Cua activity' — are exactly what a user would say; file extensions are N/A for this skill.	5 / 5
Distinctiveness Conflict Risk	Clear niche — native GUI automation via accessibility-tree snapshotting and the cua-driver tool surface — with distinct triggers and minimal overlap with other skills.	5 / 5
	Total	20 / 20 Passed

Validation

75%

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 — 12 / 16 Passed

Validation for skill structure

Criteria	Description	Result
skill_md_line_count	SKILL.md is long (1124 lines); consider splitting into references/ and linking	Warning
metadata_version	'metadata.version' is missing	Warning
metadata_field	'metadata' should map string keys to string values	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	12 / 16 Passed

Repository: trycua/cua
Path: libs/cua-driver/rust/Skills/cua-driver/SKILL.md
Commit: 359fad5

Reviewed: 4 days 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.