Interactor examples - #3
Open
Hassan-Kittany wants to merge 62 commits into
Open
Conversation
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
… option-aware self-check, robust html extraction
- apply/discard endpoints now process each path independently and always return 200 with {results:[{path,ok,error?}]}
- app.js applyOrDiscard only clears state.drafts for ok:true paths, reports failures in fixStatus
- /api/fix reads each path defensively, collecting readFailures separately, never aborts whole batch
- fixFile clean check now respects optionIds: convertCustomEffect/removeExtraJs requested but still present => needsReview
- extractHtml regex changed from anchored to non-anchored, extracts first fenced block even with surrounding prose
- Added 4 new tests (server partial batch, fix option-aware needsReview, agent prose-fence, agent no-fence trim)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace the @wix/interact Agent SDK call with a spawn of the local `claude -p --output-format json` CLI (reuses `claude login`, no API key), matching the proven interact-xp playground pattern. Parse the result from the message array. Verified end-to-end live (1.79.0 -> 2.4.0 rewrite). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Minimal, tight visual system: SF system font, neutral grays + single blue accent, translucent app bar, segmented tabs, status dots, restyled checkboxes, stat chips, thin scrollbars. Add a file filter box and full HTML escaping in the diff/list rendering. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The server runs with root = repo root, so drafts land in <repo>/.drafts/, not validator/.drafts/. Ignore the root location so generated drafts are never accidentally committed. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Redesign to explorer.html's dark floating-glass aesthetic (translucent blurred panels over a full-bleed preview, Inter, rounded corners). - Stream /api/fix per-file results via SSE; UI shows a live list with spinners → ✓/⚠/✗, a done/total count and an elapsed timer (no more opaque "fixing…"). JSON path preserved for non-streaming callers. - Add a Compare tab: Original vs Draft rendered side by side. - Tooltips on the scan summary chips explaining each category. - Rename Apply/Discard buttons to "Apply selected" / "Discard selected". Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replace the single tab row / side-by-side compare with two independent toggle groups: mode (Preview | Code | Diff) and version (Current | Draft). Toggle either to navigate — e.g. pick Code then flip Current<->Draft to compare source, or Preview then flip to compare renders. The version group hides for Diff (which inherently compares both). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- A #placeholder { display:flex } rule overrode the [hidden] attribute, so
the "Select a file" overlay stayed on top of the viewport and swallowed
pointer events from the animation. Add [hidden]{display:none!important}.
- Code/Diff were full-bleed (inset:0), starting under the left panel. Inset
them into the central column as a floating glass panel.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The updateVersion fragment and a new system-prompt VERSION RULE now require every @wix/interact import to be pinned to the exact latest version — an unpinned esm.sh import was slipping through and tripping the version check. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…S flags Replace the single category dot with independent indicators: a version dot (green=latest, yellow=old version/syntax, red=no interact) plus additive flags — a purple dot for customEffect usage and a blue "JS" badge for JavaScript animation not tied to a customEffect. A file can show several (e.g. yellow + purple, or yellow + JS). Summary chips and tooltips updated to match. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a deterministic codemod pass (codemod.js) for the mechanical fixes — pin @wix/interact@2.4.0, rename wix-interact-element→interact-element, fix the useCutsomElement typo, and rename range-offset type→unit (unit-guarded). fixFile now runs codemods first and only calls the agent when there's real semantic work: a semantic option (convert customEffect / remove JS / convert to interact), a custom prompt, or leftover structural markers (params.type/ method play-mode) after the codemod. Version-only / trivial-rename fixes are now instant, free, and deterministic — no LLM. Each result reports whether it was done via 'script' or 'agent', shown per file in the live progress list. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The old <wix-interact-element> binds via data-wix-path; the public <interact-element> binds via data-interact-key. The tag-rename codemod left data-wix-path in place, so migrated elements had no key and bound to nothing — the animation silently died (CardSpread). The codemod now renames data-wix-path → data-interact-key alongside the tag, and detect.js flags a lingering data-wix-path as an old-syntax marker so the self-check catches it. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Bump target to 2.5.1 and require the /web subpath (exports the <interact-element> custom element). The pin codemod now normalizes any @wix/interact import — versioned, unpinned, or with another subpath — to exactly @wix/interact@2.5.1/web, and uses a lookbehind so prose mentions in comments are left alone. Tests updated to the new target. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
runAgent now drives `claude --output-format stream-json --include-partial- messages`, forwarding text/thinking deltas via an onDelta callback. fixFile threads it as onLog(path,text,kind); the /api/fix SSE stream emits per-token `log` events so the UI can show the model working live. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- File list is now a collapsible folder tree (folder names + nested files) instead of full slash-paths; folders toggle open/closed, filter auto-expands. - Edge toggles collapse the left/right glass panels off-screen for an unobstructed view of the animation. - "Agent activity" button opens a modal that live-prints the model's streamed reasoning/output per file (CLI-style), with a file picker; mechanical codemod fixes show a note explaining they produced no LLM output. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Add a folder icon to each folder row in the file tree. - The agent-activity modal now distinguishes "waiting for the model to stream" (during an in-flight run) from "no agent output" (idle), and hints to restart a stale server if nothing streams — instead of always showing the misleading mechanical-fix note. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
extractHtml now clamps output to the <!DOCTYPE html>/<html> … </html>
boundaries, dropping any preamble ("per the output contract, here it is:")
or trailing remarks the model sometimes adds — which were leaking into the
draft and rendering in the preview. Fragments without those markers are
unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a /api/convert flow that runs the convert-interact-demo-example skill headlessly (SKILL.md + bundled exemplar as the system prompt) over selected files and writes each structured guideline to "Ani-Mate Prompts/<path>.md" (mirroring the source tree). Streams per-token log + per-file result over SSE like /api/fix. New endpoints: GET /api/prompts (list) and GET /api/prompt (read). Path-safe, tested. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Add Examples ↔ Prompts view tabs under the search box. Examples shows the animation tree (as before); Prompts shows the generated guideline tree from "Ani-Mate Prompts". - "Convert to prompt" button runs the selected examples through the skill with the same live streaming progress + agent-activity modal. - Selecting a prompt renders it as formatted markdown (mode tabs become Rendered / Raw); includes a small dependency-free markdown renderer. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- The agent-activity modal only auto-scrolls to the bottom when you're already there, so you can scroll up and read while it streams. - Up/Down arrows move the selection through the visible tree rows (examples or prompts), skipping collapsed folders, and scroll the row into view. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
6 tasks: vendor build (renderer+schema, GATE) → playground client → loop history store → guideline refiner → loop endpoints → loop UI. TDD on the pure/backend units; live playground + render covered by a manual smoke. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… schema Bundles @wix/interact-experience-renderer into a browser ESM module and emits the canonical experience JSON Schema, both read directly from the interact-xp source tree (read-only) via esbuild. Required aliasing `@wix/interact-experience` to its `src/index.ts` since that workspace package's `dist/` (referenced by its package.json `exports`) is never built in this checkout — mirrors the alias interact-xp's own Vite configs already use.
…ize) + serve vendor
…, refine, rounds)
… config Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…viewRound sections, refine error, stale grid) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…s, redesigned feedback dock - listSections now returns raw section.html for rendering and the sanitized copy as promptHtml for the model (mirrors the playground app) — previews no longer show redacted text/image placeholders - loop previews: one full-width card per section, stacked, tall iframes - section picker docks sticky at top; feedback (score | notes | actions) docks sticky at bottom; empty/generating states with spinner Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…iew on pill select - clicking a section pill now renders its original (static) layout in the grid before generating — /api/playground/sections returns raw html+css, and buildRenderDoc renders statically when no config is given - each preview cell has an expand button opening a full-screen overlay (Esc / backdrop / ✕ to close); cells are tagged 'original' vs 'animated' Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ing modal button - Refine/Generate buttons show an inline spinner + disable while the agent runs - status line reports 'Refining the guideline…' → 'refined — regenerating…' - new 'Agent thinking' button in the loop bar pulses while streaming and opens the live reasoning modal (refine already streams to appendLog); modal follows the active stream Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…rrent, prompt diff - sticky loop header shows 'Round N' (or amber 'Viewing round K of N') with an inline status line; loop messages no longer land in the far-away right panel - rounds rail gains a 'Current · round N' row; viewing a past round snapshots the live state and 'Back to current →' restores it - 'Δ Prompt diff' modal diffs the original .md against the working guideline (or the viewed round's) via new GET /api/loop/diff (+ tests) - consistency: refine/generate/section picks locked while viewing history or mid-run; Esc closes any open modal; rollback message says what it did Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…iff data loss - topbar agent bar: model select (CLI default/opus/sonnet/haiku) applied to every agent call (fix, convert, refine, playground generate via provider+ model in the request); context chip shows last-call tokens as % of the window + session totals with a reset button (GET/POST /api/agent/*) - agent.js captures the CLI resolved model + usage from the stream - prompt diff root cause: rounds only stored the guideline they ran WITH, so a rollback silently destroyed refined outputs and diffs showed no changes. recordRound now stores the refined output; round diffs use roundRefined() (with a legacy fallback); rollback needs a second confirming click; the diff modal explains an identical result instead of looking broken, and surfaces errors Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Pure core of the autonomous refinement engine: decide() implements the stop rule (green at threshold, amber on plateau/cap — checking cap before plateau so a stall that coincides with hitting the iteration cap reports 'cap' rather than 'plateau'), historyBlock() renders compact per-iteration summaries, and extractTriggers() pulls unique trigger types out of example HTML for judge context.
Adds validator/lib/capture.js: scrollPositions (pure), makeGif (gifenc + fast-png), and captureSweep (headless chromium scroll sweep -> PNG frames + animated GIF). Reused Chromium download was stale (build 1208 cached, 1228 required by playwright 1.61.1); ran `npx playwright install chromium` once to bring it current so the real-browser smoke test actually executes.
…d-tool agent calls
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…oute, boot recovery Wires the Task 6 createRefinery engine into server.js with real deps (listSections, generate, captureSweep, judgeIteration, refineGuideline), mounts /runs and /repo statics, adds all /api/refinery/* endpoints plus GET /render/:jobId/:iter/:sectionId, and runs markInterrupted() on boot to recover jobs orphaned by a prior process death. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…t, job view, approve flow Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…bound port to capture Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…job toggle 1. Delete a job (button in the job bar, two-step confirm) → removes the record + all frames/gifs via new POST /api/refinery/delete + deleteJob store fn (guarded against running/queued); the prompt drops back to a fresh start. 2. Fix preview flashing: renderJobView now rebuilds #jobBody only when the viewed iteration's content key changes, so 4s polls / SSE refreshes no longer recreate the <iframe>s (which reloaded and flashed). Same guard on #jobBar preserves the notes textarea mid-typing; subscribeJob no longer tears down and reopens the SSE on every poll. 3. Prompt↔job toggle: a 'Refinery' mode tab appears when a prompt has jobs; Prompt/Raw show the guideline md, Refinery shows the job view — no longer force-locked into the job view with no way back. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
/api/refinery/diff now also returns a steps array — one diff per iteration
that produced a refinement (guideline to refined = next iterations guideline);
stopping iterations (refined=null) contribute none. The diff modal renders the
overall original-to-final diff, then a captioned section per iteration
("Iteration N to N+1") so you can see exactly what each round changed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.