computer-use

Drive the user's desktop in the background — clicking, typing, scrolling, dragging — without stealing the cursor, keyboard focus, or switching virtual desktops / Spaces. Cross-platform: macOS, Windows, Linux. Works with any tool-capable model. Load this skill whenever the `computer_use` tool is available.

Quality

83%

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

85%

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 executable tool calls, a clear sequenced workflow, robust error-recovery feedback loops, and well-structured progressive disclosure into the cua-driver pack. Its main weakness is verbosity: the background-first premise and the escalation ladder are restated across several sections and could be consolidated to respect the context budget.

Suggestions

Consolidate the escalation-ladder material: it currently appears in the workflow section, the actions table note, a full dedicated section, AND the failure-mode 'Click had no effect' row — keep one authoritative version and cross-reference it to trim several paragraphs.

Reduce repetition of the 'background / no focus steal / no virtual-desktop switch' premise, which is stated in the intro, the 'Background rules' section, and again in safety/escalation prose; state it once prominently and refer back.

Consider moving the per-platform key-shortcut table into the referenced LINUX.md / MACOS.md / WINDOWS.md deep dives, since it duplicates platform-specific detail the pack already maintains.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient and dense with actionable specifics, but it runs long with repeated explanation of the verify/escalate ladder across the workflow, actions table, dedicated section, and failure-mode table, and re-states the 'background, no focus steal' premise multiple times — it could be tightened without losing clarity.	2 / 3
Actionability	Provides concrete, copy-paste-ready tool calls (`computer_use(action="capture", mode="som", app="…")`, `click`, `drag`, `scroll`, etc.), real returned-structure examples, and specific commands like `hermes computer-use doctor`, meeting the score-3 'fully executable' anchor.	3 / 3
Workflow Clarity	The canonical workflow (capture → click by index → verify) is explicitly sequenced, and the verify→escalate ladder plus the failure-mode table give explicit feedback loops (read the structured verdict, re-capture, climb rungs), satisfying the validation-checkpoint requirement even for state-changing operations.	3 / 3
Progressive Disclosure	SKILL.md is an overview focused on the Hermes-side vocabulary and clearly signals one-level-deep references into the cua-driver skill pack (MACOS.md, WINDOWS.md, LINUX.md, etc.) via an explicit install command and a listed table of what each file contains, with no nested/2+-level indirection.	3 / 3
	Total	11 / 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 is strong: it names concrete actions, gives explicit load-when guidance, and carves out a distinct niche (background, cross-platform, any-model desktop driving). Its only weak spot is trigger-term coverage, which leans on the tool name and technical phrasing rather than the natural words a user would say.

Suggestions

Add user-natural trigger terms a person would actually say, e.g. 'control my desktop', 'click buttons in an app for me', 'automate a native app', or 'drive my GUI', so the skill surfaces for everyday requests rather than only when the user references the `computer_use` tool by name.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'clicking, typing, scrolling, dragging' — and names a specific background/non-stealing constraint, matching the score-3 anchor that lists several specific concrete actions.	3 / 3
Completeness	Clearly answers 'what' (background desktop driving across platforms) AND 'when' with an explicit trigger clause: 'Load this skill whenever the `computer_use` tool is available', meeting the score-3 anchor.	3 / 3
Trigger Term Quality	Includes relevant terms like 'desktop', 'cursor', 'keyboard focus', 'virtual desktops', and the tool name `computer_use`, but the natural user-side trigger phrasing is thin ('drive the user's desktop') and lacks common variations users would actually say.	2 / 3
Distinctiveness Conflict Risk	Has a clear niche — background (non-focus-stealing) desktop GUI driving tied to a specific tool name — with distinct triggers that make it unlikely to fire for the wrong skill.	3 / 3
	Total	11 / 12 Passed

Validation

81%

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 — 13 / 16 Passed

Validation for skill structure

Criteria	Description	Result
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	13 / 16 Passed

Repository: NousResearch/hermes-agent
Path: skills/computer-use/SKILL.md
Commit: 5be99b6

Reviewed: about 15 hours 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.