CtrlK
BlogDocsLog inGet started
Tessl Logo

talk-marsden-agent-desktops

Use when the user asks about Luke Marsden's talk "Giving Every Agent Its Own Desktop: Lessons from Dogfooding HelixML" — including questions about HelixML, giving each agent its own GPU-accelerated desktop, spec-driven development with plan/implement phases, scaling agents by task vs by org-shape, centralized vs per-developer agent infrastructure, forking Zed for remote control, ZFS-cloned Docker-in-Docker dev environments, mixing local models (Llama 3.1) with frontier models (Claude Opus), the "snake eating its own tail" dogfooding approach, self-improving companies, or applying his design opinions to your own agent platform.

70

Quality

86%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

85%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.

An exemplar reference-skill body: actionable, well-sequenced with explicit verification gates, and cleanly structured around one-level-deep source files. The only meaningful weakness is repetition of the same read-outline/read-transcript/quote/cite workflow across five sub-scenarios, which inflates token cost.

Suggestions

Factor the shared procedure (read outline.md → read transcript.md → quote safe excerpts → cite line numbers) into a single 'Standard lookup workflow' block referenced once, then have each sub-scenario state only its scenario-specific deviations.

Consider merging the closely related 'Factual Q&A' and 'Teach/explain' sections, which share nearly identical steps, to further reduce redundancy.

DimensionReasoningScore

Conciseness

The body is well-organized and avoids explaining concepts Claude already knows, but the same core workflow (read outline.md → read transcript.md → quote safe excerpts → cite line numbers) is restated across five sub-scenario sections, which is more than minor redundancy and could be factored into a single shared procedure.

3 / 5

Actionability

Guidance is concrete and executable for an instruction-only skill: exact files to read ('read outline.md to locate the relevant section, then read that section of transcript.md'), exact fallback phrasing ('say "the talk doesn't address this"'), and a named ordered walk through seven audit dimensions — fully actionable without code.

5 / 5

Workflow Clarity

Each procedure is a clear numbered sequence with explicit verification gates functioning as checkpoints ('If a claim isn't in transcript.md, say "the talk doesn't address this"', 'If the framework genuinely doesn't fit... say so', per-dimension covered/partial/missing verdicts), matching the anchor for clear sequences with explicit validation and feedback loops.

5 / 5

Progressive Disclosure

SKILL.md acts as a concise overview pointing to well-signaled, one-level-deep references ('read outline.md', 'check quote.md first', 'read that section of transcript.md') organized for discovery; no nested reference chains, and content is appropriately split between the overview and the referenced source files. (No bundle files were provided in this evaluation environment, so scoring reflects the body's reference structure and signaling.)

5 / 5

Total

18

/

20

Passed

Description

87%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 highly specific, well-triggered description for a single-talk reference skill, with an explicit 'Use when...' clause and comprehensive topic enumeration. It is somewhat verbose as one long packed sentence, which keeps specificity and trigger-term quality just below the top anchor rather than at it.

Suggestions

Trim the most niche jargon phrases (e.g. 'snake eating its own tail', 'ZFS-cloned Docker-in-Docker') from the description to reduce verbosity and keep the most user-natural trigger terms.

Lead with a brief explicit action verb (e.g. 'Answer questions about...' or 'Apply frameworks from...') so the 'what Claude does' is as concrete as the 'when'.

DimensionReasoningScore

Specificity

Lists many concrete, specific topics ('GPU-accelerated desktop', 'spec-driven development with plan/implement phases', 'ZFS-cloned Docker-in-Docker dev environments', 'forking Zed for remote control') giving comprehensive subject coverage, but the actions Claude performs (answer/apply) are implicit rather than enumerated concrete verbs, so it sits just below the top anchor.

4 / 5

Completeness

Explicitly answers 'when' with a clear 'Use when the user asks about...' clause and answers 'what' by enumerating the talk's subject areas plus an action ('applying his design opinions to your own agent platform'), matching the anchor that requires both what and when with concrete trigger phrases.

5 / 5

Trigger Term Quality

Strong natural-keyword coverage ('Use when the user asks about Luke Marsden's talk', 'spec-driven development', 'scaling agents by task vs by org-shape', 'self-improving companies') that users would plausibly say, though some enumerated terms ('snake eating its own tail', 'ZFS-cloned Docker-in-Docker') are niche jargon rather than common phrasings, leaving it just short of comprehensive.

4 / 5

Distinctiveness Conflict Risk

Targets a single named talk by a named speaker with highly specific triggers (HelixML, the talk title, Zed forking, ZFS-cloned environments), creating a clear niche with minimal overlap risk against other skills.

5 / 5

Total

18

/

20

Passed

Validation

93%

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

Validation15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

metadata_version

'metadata.version' is missing

Warning

Total

15

/

16

Passed

Repository
jscraik/Agent-Skills
Reviewed

Table of Contents

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.