harbor-daytona

Use Harbor's Daytona sandbox platform for computer use — creating sandboxes, taking screenshots, sending mouse/keyboard input, and building agent loops. Use when the user wants to interact with a GUI, automate a desktop, do computer use, control a browser visually, or run Claude computer use against a Daytona sandbox.

Quality

83%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Medium

Suggest reviewing before use

Quality

Content

72%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 highly actionable, token-efficient API reference whose commands are copy-paste ready and whose agent loop is clearly sequenced. The main weaknesses are destructive troubleshooting commands (docker system prune, rm -rf of the Daytona database) presented without validation or warnings, and a monolithic inline API reference that has no bundle files to offload detail into.

Suggestions

Add validation checkpoints and warnings around the destructive troubleshooting commands: verify disk usage actually exceeds ~80% before running 'docker system prune -f', and warn that 'rm -rf services/daytona/data/db' permanently deletes all Daytona state before suggesting it.

Split the endpoint-by-endpoint reference (mouse, keyboard, display, process management) into a references/api.md file and keep SKILL.md as a quick-start plus agent-loop overview with clearly signaled links.

Deduplicate the agent loop pattern by referencing the earlier Auth and Create-a-sandbox sections instead of repeating the API_KEY setup and full curl POST body.

Dimension	Reasoning	Score
Conciseness	Nearly every line is an executable curl command or an essential fact; there is no padding explaining concepts Claude already knows. Minor inefficiencies: the Agent Loop Pattern repeats the auth setup and sandbox-creation calls from earlier sections, and the one-paragraph platform intro could be trimmed.	4 / 5
Actionability	Fully executable, copy-paste-ready curl commands with real JSON bodies, response shapes documented as comments ('returns {"screenshot": "<base64 PNG>"...}'), decode-and-save snippets, key-name tables, and a ports table — specific examples cover the common cases.	5 / 5
Workflow Clarity	The agent loop is well sequenced (create → poll until 'started' → start desktop → screenshot/act → cleanup) with a state-polling checkpoint, but the troubleshooting section issues destructive commands — 'docker system prune -f' and 'rm -rf services/daytona/data/db' — with no validation step or warning about irreversibility, which caps this dimension at 3 per the destructive-operations guideline.	3 / 5
Progressive Disclosure	No bundle files exist (no references/, scripts/, or assets/), so the entire API surface — mouse, keyboard, display, process, screenshot endpoints — is inlined as roughly 300 lines of API reference in SKILL.md. Section headers make it navigable, but this matches the anchor-3 pattern of '200 lines of API reference that could be in a separate file' rather than a well-split overview pointing to one-level-deep references.	3 / 5
	Total	15 / 20 Passed

Description

95%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 capability list in third person, an explicit 'Use when...' clause with natural trigger phrasings, and a clearly distinct niche. Only minor coverage gaps (e.g., process management, non-GUI command execution) keep specificity from the top anchor.

Dimension	Reasoning	Score
Specificity	Names the domain ('Harbor's Daytona sandbox platform for computer use') and lists several concrete actions — 'creating sandboxes, taking screenshots, sending mouse/keyboard input, and building agent loops'. Not quite anchor 5 because capabilities like process management and non-GUI command execution are absent, but it clearly exceeds the '1-2 concrete actions' of anchor 3.	4 / 5
Completeness	Explicitly answers both: what — 'creating sandboxes, taking screenshots, sending mouse/keyboard input, and building agent loops'; when — a literal 'Use when the user wants to...' clause with concrete trigger phrases. Matches the anchor-5 example structure exactly.	5 / 5
Trigger Term Quality	Comprehensive natural-language triggers: 'interact with a GUI, automate a desktop, do computer use, control a browser visually, or run Claude computer use against a Daytona sandbox' — these are exactly the phrasings a user would naturally say, with synonyms and variations of computer use all covered.	5 / 5
Distinctiveness Conflict Risk	The Daytona/Harbor computer-use sandbox is a clear niche with distinct, specific triggers ('Daytona sandbox', 'computer use', 'control a browser visually'); minimal overlap risk with generic GUI or container skills.	5 / 5
	Total	19 / 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
allowed_tools_field	'allowed-tools' contains unusual tool name(s)	Warning

	Total	15 / 16 Passed

Repository: av/harbor
Path: skills/harbor-daytona/SKILL.md
Commit: e38ae4c

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.