VisionForge is a local-first Tauri 2 desktop app for prompt-engineering and image-generation workflow orchestration using a React/TypeScript frontend and a Rust/Tauri backend. It is designed to work with local Ollama and ComfyUI services.
This repository is not marked release-ready. The May 27, 2026 proof packet reports frontend build, Rust check/test/clippy, audit, and static P31 validations passing, but runtime smoke tests against live Ollama and ComfyUI were not completed in that recorded pass. Its npm run tauri build attempt produced the release binary plus deb/rpm bundles, then failed during AppImage bundling at linuxdeploy.
Those historical receipts are tied to the snapshot named in the packet, including its then-dirty worktree. They are not fresh test results for the current checkout.
- Node.js compatible with the locked Vite 7 dependency (Node 20.19+ or 22.12+)
- Rust and the native Tauri 2 system dependencies for the target OS
- Running Ollama and ComfyUI services, with the intended models/checkpoints installed, for live workflows
npm ci
npm run build
npm run typecheck
npm audit --audit-level=moderate
cd src-tauri
cargo fmt --all -- --check
cargo check --all-targets
cargo test --all-targets
cargo clippy --all-targets -- -D warnings
cd ..
npm run tauri build
python3 codex/validation/run_all_validations.py --repo .For local development:
npm run devFor the desktop app:
npm run tauri devsrc/: React/TypeScript frontend.src-tauri/: canonical Rust/Tauri backend.docs/SPEC.md: product intent, not implementation proof.VISIONFORGE_PROOF_PACKET.md: dated build/test and claim evidence from May 27, 2026.docs/codex-runs/archive/unclassified/20260527T212459Z/files/docs/source-quarantine/P31_duplicate_backend/s/: archived copy of the quarantined P31 duplicate backend tree.
| Feature | Status | Evidence | Limitation |
|---|---|---|---|
| Single backend root | Implemented/proven | vf_assert_no_duplicate_tauri_root.py passes |
Quarantined duplicate remains under docs for audit only |
| Shared frontend config provider | Implemented/proven by static validation/build | src/context/ConfigContext.tsx, npm run build, config-provider validation |
Manual settings-to-prompt-studio smoke not run |
| Auto-approve queue handoff | Implemented/proven by static validation/build | PromptStudio auto-enqueues and queue/gallery metadata is wired |
Live pipeline smoke not run |
| Selected concept metadata | Partial | Queue payload and gallery row carry selectedConcept/selected_concept |
UI selection is lineage metadata/inspection, not a prompt-regeneration control |
| Queue/gallery metadata | Reported by the dated tests | May 27 packet reports 181 Rust tests passed | Live ComfyUI generation smoke not run |
| Ollama/ComfyUI HTTP status honesty | Implemented/proven by static validation | Status helpers check non-2xx for configured routes | Non-2xx mock tests still need expansion |
| Lock-order hardening | Implemented/proven by static validation | vf_assert_no_lock_order_inversions.py passes |
Static heuristic only |
| Export filename validation | Implemented/proven by static validation/tests | Export validation script and Rust tests pass | Invalid-row quarantine policy is fail-fast |
| File/DB rollback | Partial | Cancellation cleanup path and permanent-delete errors are explicit | Full staged-file transaction tests remain needed |
| Tauri packaging | Partial | Binary, deb, rpm built during npm run tauri build |
AppImage bundling failed at linuxdeploy |
- Release readiness is blocked until live Ollama and ComfyUI smoke tests are recorded.
- The recorded May 27 AppImage build failed at
linuxdeploy; that run created deb/rpm artifacts first. Re-run packaging before claiming the blocker is fixed or still reproducible on a new host. - Frontend lint and unit-test runners are not installed;
npm run lintandnpm run testare explicit deferred placeholders. - The selected concept UI records lineage metadata. It does not rerun prompt engineering when changed after pipeline completion.
- File/DB persistence has cleanup safeguards, but a fuller staged-file transaction test suite is still needed.
See VISIONFORGE_PROOF_PACKET.md and the P31 audit for historical command receipts, validation results, blockers, and rollback notes. The feature table summarizes that recorded proof; regenerate the checks and live-service smoke evidence for a new release decision.