CtrlK
BlogDocsLog inGet started
Tessl Logo
PluginPublic

adobe/stardust

Redesign an existing website to make it better. Built on top of impeccable.

Quality

88%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

Overview
Quality
Evals
Security
Files

phase-4-and-5-iterate-and-approve.mdskills/prototype/reference/

stardust:prototype: phases 4 and 5, iterate and approve

The text of this part of skills/prototype/SKILL.md, which keeps the heading and points here. "Above" and "below" refer to the order of the sections in that file.

Phase 4 — Open and iterate

(Phase numbering skips 3; the former Phase 3 — Compose the viewer — was removed when the per-page before/after viewer was dropped. Cross-references throughout the docs still name Phases 4, 5, 5.5.)

  1. Open the just-written or just-updated <slug>-proposed.html (or the user-chosen variant suffix when N > 1) in the browser using the open <vfs-path> shell command. This routes VFS paths through the preview service worker — do not use playwright-cli open for local prototype files, as it bypasses the preview service worker and produces a FILE NOT FOUND error. Skip in pipeline-automation mode.
  2. Mark the page prototyped in state.json — gated on the Phase 2.5 critique + Phase 2.6 audit + Phase 2.7 adapt + Phase 2.8 motion validation (when fired) result (Discipline 9). If any of the four returned ≥ 1 P0 or P1 finding (after the brand-faithful inversion auto-dismiss) and the user has not acknowledged, the page stays directed (not prototyped); surface the findings in the report grouped by source (critique: / audit: / adapt: / motion:) and recommend either fixing the issue or acknowledging explicitly. The transition itself does not require approval (a separate later step) — but it does require all gates to clear, since shipping a prototyped flag on work that fails P0/P1 on any gate misleads downstream consumers (migrate, the dashboard) about the prototype's quality.
  3. Report the prototype path and stop. Iteration happens via chat-driven impeccable commands or direct invocation (see § Iteration paths below).

Iteration paths

Refinement after the initial render takes one of two forms. They are not mutually exclusive — a single page can move through both across its lifetime.

  1. Chat-driven (default). The user gives a refinement phrase in chat — "make the hero bolder for home", "tighten the cup-note grid", "less corporate". The agent:

    • Reads the phrase against skills/stardust/reference/intent-dimensions.md to identify which axes it moves.
    • Consults skills/stardust/reference/impeccable-command-map.md to pick the matching impeccable command (often bolder, quieter, distill, typeset, colorize, or layout).
    • Shows the resolved plan to the user before executing.
    • Runs the chosen command against <slug>-proposed.html (or a specific section within it, when the phrase scopes one).
    • Re-validates per Phase 2 (:root block, data attributes, anti-toolbox audit clean, impeccable hard rules) and updates the proposed file's provenance.
  2. Direct impeccable invocation. The user runs an impeccable command directly — $impeccable bolder stardust/prototypes/home-proposed.html. Stardust isn't in the loop; the browser tab reloads whatever's on disk. This is fine and documented as a supported escape hatch. (Includes $impeccable live against the proposed file when the user wants in-browser picker iteration — that's an external tool they invoke directly; stardust does not drive its poll loop.)

The "open and reasoned" principle from the master skill applies to path 1: the agent reasons publicly about the phrase before running any command, and never silently maps a refinement to a fixed command.

Phase 5 — Approval (+ fold-back, Part III)

Approval is explicit. Stardust does not auto-approve.

The user signals approval by saying "approve " or simply "approve" in the conversation. The agent confirms the slug before writing state, then proceeds.

On approval:

  1. Verify the proposed file's provenance block lists the current active direction.md (defensive check — if the direction changed during iteration, the user must re-prototype against the new direction first).
  2. Run approval fold-back per reference/approval-fold-back.md (Part III of the merged spec). When the approved variant is non-A AND ia-fidelity is reimagined, the prototype reads the approved file's _provenance.differentiationFromA[andB] block plus the brief's compositionDelta, diffs against the active direction, and proposes folding any structural moves. The user picks fold site-wide (default) / fold page-local / don't fold. Under ia-fidelity: verbatim the fold-back is a no-op by construction (A1/A2/A3 are surface forks; nothing structural to fold). Flags: --auto-fold skips the gate; --no-fold opts out entirely.
  3. Mark the page approved in state.json. Append a { status: "approved", at: <ts> } history entry.
  4. Clear any stale flag on the page.
  5. Print:
    home: approved
      proposed: stardust/prototypes/home-proposed.html
      mobile:   adapted at Phase 2.7 (4 @media rules at 640/768/1024/1280)
      fold-back: page-local (substrate-keyed-zine pattern stays on /beers)
    
    Next: $stardust migrate home  (write final redesigned static HTML)

Adapt no longer runs at approval time — it ran at Phase 2.7 as a pre-prototyped gate. The mobile: line reports the Phase 2.7 result for review continuity.

If multiple pages are in flight, approval is per-page; the user can approve some and continue iterating on others.

skills

README.md

tile.json