<< All versions

Skill v1.0.0

currentLLM-judged scan90/100
joao-aires/agent-skills-public/improve
──Details
PublishedOctober 21, 2026 at 03:48 PM
Content Hashsha256:1659f5d472b30770...
Git SHA768a93096fb1
──Files
Files (1 file, 2.8 KB)
SKILL.md2.8 KBactive
SKILL.md · 43 lines · 2.8 KB

version: "1.0.0" name: improve description: Assess codebase improvement opportunities or implement a specified improvement objective. Use for architecture, performance, reliability, testability, UX consistency or developer-experience work; bare invocation discovers prioritized evidence-backed opportunities. user-invocable: true disable-model-invocation: true metadata: scope: generic invocation: user role: task


Improve

Read task guidance. Read deliver-reliable-changes for baseline and verification, git-contributions for implementation when requested. Load the optional stack/domain guidance selected by task guidance only when relevant and available.

Assess opportunities

Use the relevant codebase and history to find a problem that matters in practice. Look for changes that repeatedly touch unrelated modules, expensive operations in an important journey, failures that are difficult to diagnose, inconsistent interactions, or slow and unreliable development checks. Follow the evidence far enough to distinguish a cause from a disliked coding style.

For each worthwhile opportunity, explain who is affected, what evidence shows the problem, and how a bounded change could improve it. State how improvement would be measured and what behavior needs protection. Compare likely value with implementation and maintenance cost; no finding or no worthwhile change is valid. A bare request returns a short prioritized set of proposals rather than rewriting unrelated areas or creating tickets automatically.

Implement an authorized improvement

A specific requested improvement can authorize bounded implementation. Keep a bare improvement assessment read-only; do not force another approval for routine choices already authorized.

  1. Establish a baseline, diagnose the cause and define the desired outcome.
  2. Make focused changes within scope. Preserve behavior through incremental refactoring and regression checks.
  3. Compare results with the baseline and report the evidence.

Choose the relevant analysis without imposing another process:

ConcernExamine and use
UXEstablish interaction intent before changing screens; account for all affected surfaces. Use available interface assessment and interface-language expertise for craft and task copy.
Usability evidenceUse available research expertise for a requested study or actual observations.
ArchitectureExamine responsibilities, dependency direction, ownership, coupling/change hotspots, seams and testability. Use codebase-design/domain-modeling where useful, without forcing deep-module vocabulary, HTML reports or interviews.

More abstractions are not inherently better. Tracker proposals/creation and live experiments follow the task's explicit scope.

All versions