The consoles on Loom and Graphite, artifacts, providers and engines, chat context

[Zarathos94]

Stacked on #61 (feat/project-modes), which is stacked on #60 (feat/weft). Merge bottom-up.

The look: Loom and Graphite, on the web and the desktop

• apz-theme (skins, palettes, backgrounds, VS Code theme import) and the shared tokens; the old look and its CSS are gone.
• Web: the prototype shell (sidebar, top bar, Settings frame, Ctrl K palette, live pill), the kit (ui/components.ts) and every screen ported to the prototype markup.
• Desktop: the kit (console/ui.rs: cards, chips, buttons, tabs, inputs, dialogs via ui::modal), the shell and every screen redrawn to match the web, with capture hooks (APZ_WIZARD, APZ_ARTIFACT, APZ_ONBOARD=<step> …).
• One backdrop shader (backdrop.wgsl, CPU twin for WebGL2): drifting fields, Loom dots, Graphite grid, grain.

Artifacts and sharing

• apz-view bodies (charts, tiles, tables, graphs, datasets, code, HTML), shares through the hub and agentplaza.io/artifact/{id}.
• HTML artifacts run in an opaque-origin sandbox served from the hub's content origin; the console page now allows that origin as a frame source.

Providers & models

• API provider instances: label, accent, enable/disable (a disabled one takes no calls), several of one kind.
• Coding engines as instances (/v1/engines): the built-in ones, ACP registry agents (/v1/acp/registry) and local ACP commands. Versions are pinned when saved (npm integrity, PyPI or archive sha256); Update all runs a dry run first. The VM honours the launch spec the hub sends with each code_engine call (ToolDecision.engine).
• Background health checks every health_interval_s (0 turns them off), published as providers.health / engines.health.
• Usage providers: CLIProxyAPI hubs (the management key stays in the vault).
• No key ever enters an engine: engines reach models only through the gateway; subscription accounts come in through a CLIProxyAPI hub added as a provider.
• Web, desktop and TUI: the Providers screen (list and detail, Add provider in three steps, the model picker with favourites); CLI agentplaza models ….

Chats

• Every chat picks its model and effort (agent chats: the agent's model; owner conversations: their own model; one message: model on the chat call).
• A context bar per chat: used against the window, split by context part, with the work item's tool calls (GET /v1/agents/{cid}/context, GET /v1/owner/conversations/{id}/context, live through the context event). Web, desktop and TUI.

The README and the terminal UI

• README: a hero and six feature cards (animated SVG, light and dark), and "See it run": the desktop app recorded (the orchestrator chat with its effort and context breakdown, Machines in 3D, Models) and agentplaza tui recorded as animated SVG (a tour, a message to the orchestrator streaming its answer, Models with a favourite). scripts/readme/ makes all of it.
• The terminal UI redesigned: the Loom palette (truecolour, 16 colours, NO_COLOR) and an animation clock; a header with a live dot and spinners, a tab bar that keeps the active tab in view, a status line with a mode chip and hints that fit, toasts; rounded panels, status dots and selected rows on every screen, stat tiles on the overview; chat with role gutters, tool lines, a streaming cursor and a thinking shimmer; a context bar that warns over 80% and 92%; a Ctrl+K command palette; the welcome and the tour as animated cards.
• APZ_TUI_RECORD=<script.json> records the TUI as an animated SVG; the desktop app's APZ_RECORDING hides the software-rendering banner on virtual displays.

Checks run locally

• cargo fmt --all -- --check, cargo clippy --workspace --all-targets --locked -D warnings, cargo test --workspace (all pass), web tsc, node test/run.mjs, the web build; the staged set verified in a clean worktree.
• Screens compared with the prototype and the web side by side (headless captures, desktop snapshots).
• The README rendered through GitHub's Markdown API and checked in Chromium; every TUI recording checked frame by frame; cargo test -p agentplaza-cli (408) and clippy in a clean worktree.

🤖 Generated with Claude Code

[github-actions]

Tests: failed

Where	Result	Passed	Failed	Time
Linux arm64	passed	669	0	10 s
Linux x86_64	failed	4000	1	153 s
macOS Intel	passed	3453	0	79 s
macOS Apple Silicon: build only (cannot run here)	passed
Windows arm64: build only (cannot run here)	passed
Windows x64	failed	3492	1	176 s

Failures

Linux x86_64: `apid::weft_review a_code_task_and_a_research_brief_go_from_claim_to_landed_through_the_review_graph`

apid: Weft started a new store in /tmp/apid-weft-review-5822-1792496973734381702/attempt-1/state/weft (owner key /tmp/apid-weft-review-5822-1792496973734381702/attempt-1/xdg-config/agentplaza/owner.key)
weft: serving on 127.0.0.1:38593
apid: Weft serves on 127.0.0.1:38593 (process)
timing op=sleep id=305 total_ms=0.4
timing op=sleep id=306 total_ms=0.4
timing op=sleep id=307 total_ms=0.4

note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

running 1 test
w-1 at root: 0 files in /tmp/apid-weft-review-5822-1792496973734381702/vms/vm300/pr1-queues
w-1.d1: 1 files from the base
w-1.d2: 2 files from the base
w-1.d3: 2 files from the base
test a_code_task_and_a_research_brief_go_from_claim_to_landed_through_the_review_graph has been running for over 60 seconds
test a_code_task_and_a_research_brief_go_from_claim_to_landed_through_the_review_graph ... FAILED

failures:

failures:
    a_code_task_and_a_research_brief_go_from_claim_to_landed_through_the_review_graph

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 144.86s



thread 'a_code_task_and_a_research_brief_go_from_claim_to_landed_through_the_review_graph' (5823) panicked at crates/apid/tests/weft_review.rs:204:9:
the review did not finish
apid: owner token created: /tmp/apid-weft-review-5822-1792496973734381702/attempt-1/state/owner-token (use it with `agentplaza hub login`)
db: no APZ_DATABASE_URL; components stay in their files; transcripts, infractions and the IDE audit go to files under the state dir
vault: initialised with a key file at /tmp/apid-weft-review-5822-1792496973734381702/attempt-1/config/vault.key
apid: listening on http://127.0.0.1:36489
apid: Weft started a new store in /tmp/apid-weft-review-5822-1792496973734381702/attempt-1/state/weft (owner key /tmp/apid-weft-review-5822-1792496973734381702/attempt-1/xdg-config/agentplaza/owner.key)
weft: serving on 127.0.0.1:38593
apid: Weft serves on 127.0.0.1:38593 (process)
timing op=sleep id=305 total_ms=0.4
timing op=sleep id=306 total_ms=0.4
timing op=sleep id=307 total_ms=0.4

note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

Windows x64: `apid::weft_review a_code_task_and_a_research_brief_go_from_claim_to_landed_through_the_review_graph`

apid: Weft started a new store in C:\Users\actions\AppData\Local\Temp\apid-weft-review-1700-1792496769136514700\attempt-1\state\weft (owner key C:\Users\actions\AppData\Local\AgentPlaza\config\owner.key)
weft: serving on 127.0.0.1:60018
apid: Weft serves on 127.0.0.1:60018 (process)
timing op=sleep id=305 total_ms=0.7
timing op=sleep id=307 total_ms=2.6
timing op=sleep id=306 total_ms=4.2

note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

running 1 test
w-1 at root: 0 files in \\?\C:\Users\actions\AppData\Local\Temp\apid-weft-review-1700-1792496769136514700\vms\vm300\pr1-queues
w-1.d1: 1 files from the base
w-1.d2: 2 files from the base
w-1.d3: 2 files from the base
test a_code_task_and_a_research_brief_go_from_claim_to_landed_through_the_review_graph has been running for over 60 seconds
test a_code_task_and_a_research_brief_go_from_claim_to_landed_through_the_review_graph ... FAILED

failures:

failures:
    a_code_task_and_a_research_brief_go_from_claim_to_landed_through_the_review_graph

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 111.21s



thread 'a_code_task_and_a_research_brief_go_from_claim_to_landed_through_the_review_graph' (3656) panicked at crates\apid\tests\weft_review.rs:204:9:
the review did not finish
apid: owner token created: C:\Users\actions\AppData\Local\Temp\apid-weft-review-1700-1792496769136514700\attempt-1\state\owner-token (use it with `agentplaza hub login`)
db: no APZ_DATABASE_URL; components stay in their files; transcripts, infractions and the IDE audit go to files under the state dir
vault: initialised with a key file at C:\Users\actions\AppData\Local\Temp\apid-weft-review-1700-1792496769136514700\attempt-1\config\vault.key
apid: listening on http://127.0.0.1:59957
apid: Weft started a new store in C:\Users\actions\AppData\Local\Temp\apid-weft-review-1700-1792496769136514700\attempt-1\state\weft (owner key C:\Users\actions\AppData\Local\AgentPlaza\config\owner.key)
weft: serving on 127.0.0.1:60018
apid: Weft serves on 127.0.0.1:60018 (process)
timing op=sleep id=305 total_ms=0.7
timing op=sleep id=307 total_ms=2.6
timing op=sleep id=306 total_ms=4.2

note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

_{Commit 8a0d9fa · the run}

[github-actions]

Lint: passed

Check	Result
rustfmt	passed
clippy (warnings fail)	passed
cargo deny (advisories, licences, sources)	passed
migrations (applied ones unchanged, new ones in order)	passed

_{Commit 8a0d9fa · the run}

[github-actions]

Web console: passed

Check	Result
type-check	passed
build (with the wasm engine)	passed
tests	passed

_{Commit 8a0d9fa · the run}

[github-actions]

Hub API (Bruno): passed

244/244 API requests pass · 158/575 operations tested · 17 fuzzing findings (reported, they do not fail the pull request)

The report

_{Commit 8a0d9fa · the run}