docs: a2ui-rs screen-addressing proposal — address the screen, not the pixels#204
Conversation
…e pixels
Operator ask (2026-07-14): Citrix-shaped remote desktop via ClassView x
WideFieldMask addressing; askama/ERB fieldview rendering from memory without
serialization; Argon2 for the rdp-2graph; A2UI fork -> a2ui-rs as the OGAR
render target, reimagined with the workspace's optimizations.
The charter (docs/A2UI-SCREEN-ADDRESSING-PROPOSAL.md):
- One sentence: down the wire (GUID 16B, mask delta, LE values), up the wire
ActionInvocations; the client holds the ClassView/template codebook — the
ClassView registry is the font of the desktop.
- Screen = document = the SAME positional projection (doc-layer A3 extended);
DocRenderer gains its 4th adapter: a2ui (live surface).
- RBAC by projection: wire content = WideFieldMask INTERSECT role-mask — an
unauthorized field is absent, not hidden. Pixels cannot make that guarantee.
- Thin-client answer: wasm = the same Rust (ClassView resolve + askama)
compiled for the browser; no bespoke thin client needed for v1.
- A2UI fork verified: upstream pattern kept (declarative intent, native
render, renderers/{react,angular,flutter,lit}); prior-art hamming.proto
REGRADED pre-V3 (service shape kept, 1250B fingerprint payload replaced by
16B key + mask + values).
- Traps: component-vocab fork / behavior-in-tree (SURREAL-AST, UI edition) /
serialization creep.
- Killer probe P-REHOST: re-render one harvested MedCare screen from
CompiledClass x ClassView x askama + fire one harvested ActionDef — the
transpile endgame (harvest the app -> re-render the app).
- Waves W0-W5; W0 gates: mint AdaWorldAPI/a2ui-rs (operator) + council.
Ledger: D-A2UI-SCREEN-ADDRESSING appended ([S]).
Co-Authored-By: Claude <noreply@anthropic.com>
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_42841671-1657-4788-92b5-836d22294842) |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b0ae858052
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| 4. **RBAC by projection (the security headline):** what leaves the server is | ||
| `WideFieldMask ∩ role-mask` (`FieldMask::intersect`, the existing RBAC | ||
| brick from the SoC/dedup pipeline). A field the role may not see is not |
There was a problem hiding this comment.
For any addressed surface with fields past 64, this guarantee is not implementable with the current RBAC mask seam: ClassRbac::field_mask still returns a narrow FieldMask (docs/CLASSID-RBAC-KEYSTONE-SPEC.md:85-90), while the wide renderer exists specifically because WideFieldMask is needed to address every position past FieldMask::MAX_FIELDS (crates/ogar-render-askama/src/rust_class.rs:154-180). Intersecting a WideFieldMask with a narrow role mask (or relying on FieldMask::intersect) either cannot express permissions for fields >=64 or risks falling back to an all-fields sentinel, so the stated “unauthorized field is absent from the wire” property can silently stop holding for the wide screens this proposal targets; the spec should require a WideFieldMask role projection or page the surface before applying RBAC.
Useful? React with 👍 / 👎.
docs(a2ui): C1.4 correction — projection security requires WIDE role masks (codex P2 on #204)
2 participants
What
The charter for the operator's 2026-07-14 ask: a Citrix-shaped remote desktop that addresses the screen instead of pushing pixels —
AdaWorldAPI/A2UIreimagined as the OGAR render targeta2ui-rs, on ClassView × WideFieldMask.docs/A2UI-SCREEN-ADDRESSING-PROPOSAL.md+ theD-A2UI-SCREEN-ADDRESSINGledger entry.The shape
(GUID key 16 B, WideFieldMask delta, changed LE values); up =ActionInvocations. No pixels, no JSON/proto in the hot path (Firewall ADR-022/023) — askama renders from borrowed memory, zero serialization (the ERB-pattern fieldview).DocRenderergains its 4th adapter — tesseract-PDF / Spire.Doc / askama / a2ui (live surface).WideFieldMask ∩ role-mask(FieldMask::intersect) — an unauthorized field is absent from the wire, not hidden client-side. Pixel streams can't make that guarantee.ogar-auth(Argon2 KDF) +ogar-encryption(transport). No new crypto.navigates_to/ActionInvocationedge in the same closed vocabulary the C# harvest emits statically.Grounded preflight (fork cloned + verified this session)
Upstream A2UI pattern confirmed ("agents speak UI";
renderers/{react,angular,flutter,lit,web_core}). Prior art in the fork regraded, not deleted:proto/a2ui/hamming.proto(bitpacked fingerprint frames, XOR deltas, BindSpace-era) is the pre-V3 carrier — its service shape survives (RenderStream / ActionStream / codebook sync), its payload is replaced by the V3 facet (16 B key + mask + values vs 1,250 B frames, ~78× smaller per addressed node, key prerenders layout with zero value decode).Traps named before they bite
T1 component-vocabulary fork (widget skins are templates, never a second closed vocabulary) · T2 behavior-in-tree (the SURREAL-AST trap, UI edition — actions are
ActionDefrefs on the Core node) · T3 serialization creep (LE end-to-end; membrane adapters only at the edge).The killer probe — P-REHOST (the transpile endgame)
Re-render one harvested MedCare screen from its
CompiledClass× ClassView × askama and fire one of its harvestedActionDefs round-trip: harvest the app → re-render the app, no WinForms. Everything needed exists (Klickwege golden v2, 162 CompiledClass / 2,748 ActionDefs,ogar-render-askama).Gates (W0 — not jumped)
Mint
AdaWorldAPI/a2ui-rs(operator) · council verifies this spec · hamming.proto regrade in the fork. Canon-free: no mints in v1 (widget skins are templates, not concepts).Generated with Claude Code
Generated by Claude Code