opencode-impeccable-audit

name	opencode-impeccable-audit
version	0.1.0
implements_spec	0.4.0
implements_gate	impeccable-audit
description	Post-implementation visual quality audit against the deployed component. Re-scores the seven design dimensions (typography, color, spatial, motion, interaction, responsive, UX writing) against the shipping code, runs the 24-item AI-slop anti-pattern scan, and blocks branch close on Red findings (per ADR-0011). Use after a UI-shipping phase's implementation is complete and the changes are visible in the dev server. May also be invoked retroactively against shipped UI to catch drift between mockup and implementation.

opencode-impeccable-audit

This skill fulfills the impeccable-audit gate from agenticapps-workflow-core/spec/02-hook-taxonomy.md and is bound by ADR-0011. It is the post-implementation half of the design quality contract; the pre-implementation half is opencode-design-critique.

When to invoke

• A phase's changeset modifies the visual surface of a shipping UI (typography, color, layout, spacing, motion), AND
• The implementation is complete (all tasks marked complete by superpowers:verification-before-completion), AND
• The change is visible in the dev server.

MAY also be invoked retroactively on shipping UI that has not been audited under this gate — surface drift between mockup intent and implementation reality is a common gap.

What this skill does

1. Boot the dev server, navigate to the surface. Same browser tooling as opencode-qa. Capture screenshots at standard viewport sizes (1280×720 baseline; 390×844 mobile for responsive surfaces).

2. Re-score the seven dimensions (same dimensions as opencode-design-critique) — this time against the SHIPPED code, not the spec or the mockup:

   • Typography, Color, Spatial, Motion, Interaction, Responsive, UX writing — each 0–100

3. Run the 24-item anti-pattern scan. Same catalog as opencode-design-critique, applied to the rendered output. Flag each occurrence with:

   • Anti-pattern name
   • Where it appears (component / file / screenshot region)
   • Severity (Red — blocks branch close / Yellow — followup OK / Green — minor noise)

4. Compare to mockup intent. If a UI-SPEC.md and / or pre-implementation critique exist, diff the shipped output against the chosen mockup variant. Note any drift.

5. Compute the composite score. Same weighted mean as design-critique. Compare to the project's quality bar (default ≥ 90 per ADR-0011, project-overridable).

6. Write the audit document to .planning/phases/<NN>-<slug>/IMPECCABLE-AUDIT.md:

   # Impeccable audit — phase {{N}}
   
   Auditor: opencode-impeccable-audit v0.1.0
   Mode: post-implementation
   Surface: {{component / route}}
   Quality bar: ≥ {{N}}
   
   ## Scores (vs. shipped code)
   
   | Dimension | Score | Notes |
   |---|---|---|
   
   ## Anti-pattern flags
   
   | Severity | Anti-pattern | Location | Fix |
   |---|---|---|---|
   
   ## Mockup-to-implementation drift
   
   - Spec'd: …
   - Shipped: …
   - Drift severity: …
   
   ## Verdict
   
   <pass | pass-with-followups | block>
   

7. Block branch close on Red findings. Per ADR-0011, any Red anti-pattern flag blocks branch close until fixed or accepted via ADR.

Required evidence (per spec/06)

• IMPECCABLE-AUDIT.md exists in the phase directory
• Scores recorded for all seven dimensions
• Anti-pattern flags enumerated with severity, location, fix
• Mockup-to-implementation drift section present (or noted as N/A if no UI-SPEC.md existed)
• Composite score and verdict explicit
• Referenced from REVIEW.md or VERIFICATION.md

Failure modes

• Auditing the spec instead of the shipped code. Pre-impl critique is opencode-design-critique. This gate is post-impl — audit what actually shipped.
• Treating Red findings as advisory. Per ADR-0011, Red blocks branch close. Accept via the database-security-acceptance ADR pattern (a generic acceptance ADR — even though that template was authored for DB findings, the pattern transfers).
• Skipping because "the design-critique was clean." Pre-impl and post-impl catch different drift. Both fire.
• Anchoring on AI-default styling. Purple gradients, Inter-everywhere, weak hierarchy from too many weights — score these honestly, not generously.

Notes for the opencode host

• This skill shares infrastructure with opencode-qa (browser tooling, dev server). Run them in sequence on a UI-shipping phase: qa → impeccable-audit. Same screenshots can be referenced from both reports.
• The 24 anti-pattern catalog ships inline with this skill until the upstream pbakaus/impeccable skill is installed; once installed, defer to it.