agentleFS
KyaniteLabsKyaniteLabs/tastecheck/skills/improve-existing-website/SKILL.md

improve-existing-website

Skill12 starsChanged 12 days agoMITView on GitHub

Use when improving an existing website or codebase while retaining its recognizable identity, especially when evidence, scope, or redesign approval is uncertain.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use references/signal-vs-drift.md when repetition, age, or inconsistency could be intentional.

    line 36
  2. Distinctive:

    Done when: every proposed change names what survives and why it is authorized.

    line 39
  3. Distinctive:

    Ask only questions that change the safe direction.

    line 45
  4. Distinctive:

    Do not solve a system problem with isolated polish.

    line 47
  5. Very specific:

    If work pauses, save current state, unresolved question, evidence, and exact next decision using references/inspection-state-machine.md.

    line 60
  6. Very specific:

    Stop / handoff: Pause when evidence is insufficient for a safe inference (+2 in contract.json).

    line 88

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Improve Existing Website

Improve what is weak without sanding away identity. Existing work contains signal and drift; evidence decides which is which.

1. Inspect the product people actually use

Open the rendered priority path before prescribing. Trace entry, decision, action, and recovery at wide/narrow widths. Inspect hierarchy, tokens, repeated components, content, supplied analytics/user evidence, and business cues.

Label what you can point to EVIDENCE and interpretations INFERRED. “The primary action changes label across three screens” is direct; “the brand values warmth” is inferred unless brief/product evidence says so.

Done when: priority path, recurring choices, and highest-cost failures are observable.

2. Separate identity from drift

Record signals worth preserving: distinctive type, color relationships, interaction patterns, voice, layout logic, or recognizable assets. Classify each change:

ClassMeaningAuthority
Preservecoherent, repeated, brief-supported identityrepair defects without replacing the signal
Normalizeaccidental variation against an established patternsafe to align when evidence is strong
Approval neededa new hierarchy, palette, voice, or material redesignshow the choice and wait for owner agreement

Use references/signal-vs-drift.md when repetition, age, or inconsistency could be intentional. Age alone is not drift evidence.

Done when: every proposed change names what survives and why it is authorized.

3. Form the smallest coherent direction

Complete assets/INFERRED-SYSTEM.template.md across audience, territory, type, color, layout, components, voice, and accessibility. Score readiness with references/inspection-state-machine.md. Ask only questions that change the safe direction; otherwise state inference and evidence.

Do not solve a system problem with isolated polish. Choose a narrow vertical slice proving the direction across a real path (navigation, content section, form/state, narrow layout) before propagating it.

Done when: the direction can be approved, rejected, or resumed without reconstructing the investigation.

4. Approve, repair, and verify

For each repair, record observed failure, preserved signal, smallest coherent edit, owner/ approval source, and proof. After approval, use references/repair-handoffs.md for specialist passes. Verify the same path/viewports; comparison must show defect changed and signal survived.

If work pauses, save current state, unresolved question, evidence, and exact next decision using references/inspection-state-machine.md. Resume from that checkpoint, not a fresh opinion.

Done when: report distinguishes repaired drift, preserved identity, browser/static proof, and deliberately deferred ambiguity.

Reference files

  • references/audit-procedure.md — extraction order.
  • references/signal-vs-drift.md — identity versus accident.
  • references/decision-records.md — scope ADR rules.
  • references/inspection-state-machine.md — readiness/approval.
  • assets/improve-checklist.md — completion ledger.
  • references/repair-handoffs.md — routing/skip rules.
  • references/interview-contract.generated.md — canonical authority.

Self-check

assets/improve-checklist.md is completion authority for inspection, labels, approval, handoffs, and evidence.

<!-- contract:v1:start -->

Contract (generated)

Canonical detail: contract.json.

  • Route: An existing site or codebase needs improvement while preserving its intended identity (+1 in contract.json); avoid: A greenfield project needs direction before anything exists to inspect (+1 in contract.json)
  • Exclude: Do not prescribe or edit before inspecting evidence (+1 in contract.json)
  • Stop / handoff: Pause when evidence is insufficient for a safe inference (+2 in contract.json); receives [none] -> sends [a11y-pass, art-direction, cognitive-a11y, color-system, component-states, data-viz, deslop-ui, empty-states, humanize-copy, i18n-ready, responsive-layout, spacing-system, tasteroll, theming, web-typography]
  • Output: An evidence-labelled inferred system, readiness decision, and scoped repair plan or implementation report
  • Evidence: table_with_evidence with status, reason, remediation, evidence, provenance.
<!-- contract:v1:end -->

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.