CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

computer-use-driven-iteration

Improve an interactive product through repeated computer-use simulations, bounded Claude implementation, verification, and reflection to find opportunities to collapse complexity. Use when asked to keep exercising and improving a UI across implementation waves. Skip one-off browser actions, routine fixes, and API-only work without an interactive product to exercise.

Quality

83%

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.

A well-structured coordination skill: the wave loop is clearly sequenced with explicit validation, stop criteria, and recovery paths, and detail is properly delegated to clearly linked sibling skills. The remaining gaps are the absence of any concrete worked example and a time-sensitive model-version mention that will age.

Suggestions

Generalize 'Claude Opus 5.5' to a capability-based designation (e.g. 'the strongest available Claude model') since the body already states responsibilities survive model-availability changes; the specific version will go stale.

Add one compact worked example — a representative journey and its matching delegation brief with stop condition — to ground the abstract directives in actionability.

Trim a few aphoristic sentences (e.g. 'Parallel work earns its place when it answers a distinct question') to their operative content to tighten token efficiency.

DimensionReasoningScore

Conciseness

The body is dense and assumes competence — no concept explanations, and nearly every sentence is directive ('A running worker's newer code is not evidence about the app currently open in the browser'). It is not anchor 5 because of the time-sensitive version mention 'Claude Opus 5.5' outside any deprecated section and a few ornamental sentences ('Parallel work earns its place when it answers a distinct question') that could be tightened.

4 / 5

Actionability

For an instruction-only skill the guidance is mostly actionable: concrete inputs for delegation ('Give Opus the relevant baseline, raw evidence, affected consumers, allowed changes, and a bounded stop condition'), explicit recording requirements ('Record the steps, expected behavior, observed behavior, and limits'), and a concrete resume-state checklist. It is not anchor 5 because no worked example (a sample journey or a sample delegation brief) is given to anchor the abstract directives.

4 / 5

Workflow Clarity

The four-phase loop (Establish the outcome and evidence → Choose and execute the next wave → Reflect and challenge → Continue or finish) is clearly sequenced with explicit validation checkpoints: 'Inspect the returned diff, run relevant checks', 'Re-exercise the changed journey against the integrated version', 'Distinguish worker-reported checks from checks Codex reproduced', explicit stop criteria, and a blocked-UI recovery path. Error-recovery loops and a checklist for resumable state are both present, matching anchor 5; anchor 4's 'minor validation gaps' is not evident.

5 / 5

Progressive Disclosure

The skill is a clean overview that pushes mechanics into five clearly signaled, one-level-deep linked skills ('The delegation skill owns clone preparation, dispatch, correction, and integration mechanics'), each with an explicit routing rule for when to use it. No bundle files exist to inline incorrectly, no nested references, and the four section headers make navigation easy — matching anchor 5's clear-overview structure.

5 / 5

Total

18

/

20

Passed

Description

76%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 with an explicit what, an explicit 'Use when' trigger clause, and unusually good negative boundary guidance. Its main weakness is trigger vocabulary dominated by process jargon ('implementation waves', 'collapse complexity') that users would not naturally say, which limits trigger-term quality.

Suggestions

Add user-natural synonyms to the trigger clause, e.g. 'Use when asked to keep exercising, testing, and improving a UI/app across implementation waves' so everyday phrasings like 'test the UI' or 'iterate on the app' match.

Translate one piece of jargon into plain language in the first sentence, e.g. replacing 'find opportunities to collapse complexity' with 'find complexity to remove', to widen natural trigger coverage without losing the niche.

DimensionReasoningScore

Specificity

The description lists several concrete process actions — 'repeated computer-use simulations, bounded Claude implementation, verification, and reflection' — rather than generic filler. It stops short of anchor 5 because 'collapse complexity' and 'implementation waves' remain abstract process language rather than fully concrete, comprehensive operations.

4 / 5

Completeness

It explicitly answers both parts: 'what' via 'Improve an interactive product through repeated computer-use simulations, bounded Claude implementation, verification, and reflection' and 'when' via the concrete trigger clause 'Use when asked to keep exercising and improving a UI across implementation waves', plus an explicit skip boundary. The 'when' is already explicit and concrete, so anchor 4's 'could be more explicit' does not apply.

5 / 5

Trigger Term Quality

Natural phrases like 'keep exercising and improving a UI' and 'one-off browser actions, routine fixes' are present, but the distinctive vocabulary ('computer-use simulations', 'bounded Claude implementation', 'implementation waves', 'collapse complexity') is internal jargon a user would not naturally say, and common synonyms like 'iterate on the app' or 'test the UI' are missing. Not anchor 4 because keyword coverage relies more on skill-specific jargon than on terms users actually utter.

3 / 5

Distinctiveness Conflict Risk

The multi-wave computer-use iteration niche with delegation is mostly distinct, and the skip clause ('Skip one-off browser actions, routine fixes, and API-only work') actively disambiguates from one-off browser and routine-fix skills. Not anchor 5 because a plain 'improve the UI' request could still plausibly trigger this heavyweight loop instead of a simpler frontend-iteration skill.

4 / 5

Total

16

/

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.

Validation — 15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

relative_links

Relative link issues: 6 suspicious

Warning

Total

15

/

16

Passed

Repository

EpicenterHQ/epicenter

Path

.agents/skills/computer-use-driven-iteration/SKILL.md

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.