[ymote]

Why

OctoSense-org/OctoScript#53 lets L0 cards show text the model wrote (a digest summary, a draft, the agent's side of an in-card chat), marked AI-written. Its three lowerings emit the mark, but the kit side was missing: kit::lower wraps model text in l0_ai_text(…), which did not exist, so a card holding model text would not evaluate. This PR adds the kit function, reads the stamp, and draws the mark.

What changes

• Kit (components/l0/_kit.octoscript): adds fn l0_ai_text(node) { node.ai = 1 return node }. Like l0_live, it only stamps the node.
• Model (octoscript-node): adds Attrs::ai. A new module, octoscript_node::ai, holds the treatment as a UiNode → UiNode rewrite (expand_ai_marks). Any backend can run the same rewrite, so they all draw the same mark. octoscript-render re-exports it.
• Evaluator (octoscript-render/src/eval.rs): reads ai.
• Renderer (octoscript-makepad): to_makepad_ui and to_makepad_l0_ui[_with_events] expand the marks before the ink pass. The beauty inspector JSON reports ai.

The mark

It follows the "AI SUMMARY" eyebrow from the OctoSense mail MVP (#261): a small sparkle and an AI eyebrow drawn above the words. Details:

• Glyph: Font Awesome solid wand-magic-sparkles (U+E2CA), drawn in the icon face. The ✦ character (U+2726) is not in any text face the kit ships (Roboto, NotoSans, Inter, LXGW, NotoColorEmoji), so it would render as tofu.
• Colour: the words' own ink at 72% alpha, so it fits every palette without adding a colour. A test runs it across all 11 palettes.
• Size: about 0.7× the body, clamped to 9–12pt, weight 600, slightly tracked.
• Placement: the mark sits next to the first text of the stamped node, not around the node. A chat bubble is stamped on its outer column and a band on its filled card. Placed this way, the mark stays on the same fill as the words, so it never ends up as ground-coloured ink on the ground.
• The words themselves: role, size, face and colour are unchanged.

Plain text only

Model text stays the text: of a plain makepad Label, emitted as an escaped string literal. There is no Markdown, HTML, TextFlow or link widget, and no autolinking. Tests check that markdown, javascript: links, HTML, an injected on_click, and l0:{…} targets all stay inert literals.

l0_ai: true (the Makepad lowering)

No change is needed. makepad's #[derive(Script)] apply reads only declared fields by name (value_for_apply(value, id!(field))), so an undeclared key on a widget object is never read. l0_ai: true is ignored exactly as l0_event:/l0_key: already are. Nothing in this repo consumes makepad::lower's output. The mark the user sees comes from the kit path above.

ChatEntry

Per #53, ChatEntry lowers to the existing bubbles: user → l0_bubble_me, model → l0_ai_text(l0_bubble_them(…)), host → l0_bubble_them, plus the *_long forms. A test renders all three roles. The user's bubble is right-aligned on the accent, the agent's and host's are left-aligned on the panel fill, and only the model's bubble is marked. The mark sits inside its bubble.

Tests

• crates/octoscript-node/src/ai.rs: 4 unit tests. They cover: untouched when unstamped; mark above the words in the same ink, quieter; idempotent; inside a bubble; text kept verbatim.

• crates/octoscript-makepad/tests/l0_ai_text.rs: 9 tests against the real kit. The kit calls are copied from the shapes Repin makepad to main f08eec53 (terminal draws Chinese text) #53's tests/ai_text.rs asserts. They cover:

  • stamping
  • every markable role, including Field and Band
  • mark order, ink and icon face
  • all palettes
  • plain-label and injection checks
  • an AI draft in a Field that stays editable
  • the three-role chat transcript
  • an unmarked card that stays unchanged

• These tests are independent of the pinned Octoscript revision. I also ran them against Repin makepad to main f08eec53 (terminal draws Chinese text) #53's head (3ca709c), along with a scratch end-to-end test (l0::prepare of Repin makepad to main f08eec53 (terminal draws Chinese text) #53's summary card and chat.card in 4 moods), and they pass.

• cargo test --workspace: all green (octoscript-makepad 115 passed, 8 ignored as before).

• cargo clippy -p octoscript-node -p octoscript-render -p octoscript-makepad --all-targets -- -D warnings: no new findings. The crate already failed on main because of 3 findings in code this PR does not touch, so this run is not fully clean:

  • clippy::needless_borrow, l0.rs:98
  • clippy::erasing_op, material.rs:672
  • an unused build_with_sys_stub, tests/l0_cards.rs

  With those three allowed, it is clean.

Downstream (OctoSense)

• Repin Octoscript-Makepad (octoscript-makepad and octoscript-node revs in Cargo.toml, and native-runtime.lock.json) to this PR's merge, together with Octoscript (octoscript-ui-l0 and friends) to Repin makepad to main f08eec53 (terminal draws Chinese text) #53's merge. The .sources checkout supplies _kit.octoscript through OCTOSENSE_WORKSPACE. This repo's own octoscript-ui-l0 pin (Cargo.toml, runtime.json) also moves to Repin makepad to main f08eec53 (terminal draws Chinese text) #53 once that merges.
• apps/appcard/app/app/src/app/l0_eval.rs builds Attrs { … } exhaustively, so it will stop compiling until it adds ai: int_prop(vm, value, id!(ai)),.
• l0_widgets::to_dsl (the device renderer) must call octoscript_node::ai::expand_ai_marks(&mut tree) on a clone before emitting. Otherwise the stamp is read but nothing is drawn. glance_card and phone already go through octoscript_makepad::to_makepad_*, which do this.

🤖 Generated with Claude Code