Takes d6:strands and d6:strands-typescript from 32/35 to 34/35.
- shared-state-read: the turn-2 fixture leg wrongly pinned turnIndex:0, so the
aimock matcher skipped it on turn 2 -> 404 -> turn-2 sse-missing. Drop
turnIndex to mirror the langgraph-python gold-standard fixture.
- multimodal: sample.png/pdf/wav shipped as git-LFS pointers, so deploy/test
environments without 'git lfs pull' served the ~130-byte pointer text as the
upload -> the run never started (runsFinished=0). Ship them as regular
binaries via a per-integration .gitattributes lfs-unset + real bytes,
mirroring langgraph-python's convention.
Remaining red (gen-ui-declarative) is a Strands A2UI-dynamic run-completion bug
(reproduces on real-LLM staging too): the surface paints but generate_a2ui
never completes, so the run hangs 'Running'. Tracked separately.
The built-in-agent showcase 400s on every prompt against the real OpenAI
Responses API. The state tools (AGUISendStateSnapshot / AGUISendStateDelta /
set_steps) declared their arbitrary payloads as z.any(), which serializes to
a typeless JSON Schema property. @tanstack/openai-base's
isStrictModeCompatible only screens for oneOf/allOf/not/$ref/$defs, so it
misses the missing type, sends the tool with strict:true, and OpenAI
rejects it:
400 Invalid schema for function 'AGUISendStateSnapshot':
In context=('properties','snapshot'), schema must have a 'type' key.
This is the real-OpenAI form of OSS-132 (the earlier zod3/zod4 drift that
produced a typeless root is already fixed by the zod 4 migration). It was
masked in production because the deployed showcase runs against aimock,
which replays fixtures without validating the request schema.
Fix: declare each arbitrary payload as a JSON-encoded string (z.string() ->
{ type: string }, strict-valid) and parse it in the server handler.
parseJson also tolerates an already-parsed object/array so recorded aimock
fixtures keep working. The TanStack->AG-UI converter reads the parsed
structure off the tool result, so it needs no change. Verified accepted and
correctly round-tripped by gpt-4o on the real Responses API with all three
tools attached.
Also bump @tanstack/ai 0.18.0 -> 0.34.0 and @tanstack/ai-openai 0.9.1 ->
0.15.4, and add the newly-required @tanstack/ai-client 0.18.2 peer. Adapter
schema output is byte-identical across the bump.
Switch declarative-gen-ui to runtime auto-injection (injectA2UITool: true, plain
agent), matching the langgraph-python / AWS Strands gold standard. The ag-ui-adk
0.7.0 adapter auto-injects generate_a2ui via plan_a2ui_injection. Verified under
the canonical D6 driver: all four declarative pills pass (sales-dashboard,
team-performance, at-risk, top-account), plus a2ui-fixed-schema and
beautiful-chat.
Remove the stale known-failing mark from the a2ui-recovery heal e2e and the
OSS-374 inner-render-disambiguation notes (fixture _meta, QA doc, agent
docstring, route comment). Verified against the aimock journal: ag-ui-adk 0.7.0
forwards the run conversation into the inner render_a2ui call, so each pill
matches its own fixture by userMessage and both heal and exhaust pass. The
recovery demo stays backend-owned (the only path with the recovery loop) and
ADK-only (OSS-375 tracks langgraph-python parity).
Bump every @copilotkit/* dependency across the showcase integrations and
the shell from 1.60.2 (and stray "latest" override pins) to an exact
1.61.1 pin, and move the canonical pin source of truth to match.
Regenerate each standalone npm package-lock.json with the same
--legacy-peer-deps flag the Dockerfiles use for "npm ci".
- showcase/integrations/*/package.json + package-lock.json
- showcase/integrations/langgraph-typescript/src/agent/*
- showcase/shell/package.json + package-lock.json
- showcase/scripts/showcase-canonical-pins.json: canonical 1.60.2 to 1.61.1
aimock stays on its own version line (1.26.1). The Python copilotkit SDK
was already 0.1.94 across every requirements.txt, so no change there.
validate-pins ratchet is unchanged (FAIL=38, identical hash);
validate-parity, validate-fixture-tool-surface, and the showcase/scripts
vitest suite (2102 tests) all pass.
Replace the "Tracked in Linear" placeholders with the filed issue IDs:
OSS-374 (showcase-aimock inner-subagent disambiguation) and OSS-375
(recovery-demo langgraph-python parity).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The `heal` e2e test fails against the live showcase aimock because the harness
cannot disambiguate the two pills' inner render_a2ui sub-agent calls (the
backend-owned middleware issues the inner call with a generic render prompt +
shared suggestion context, so the "last user turn" aimock keys on is not the
pill prompt). Both pills match the same inner fixture (the EXHAUST one, by
first-match order), so the heal pill exhausts instead of healing.
This is an aimock harness limitation, not a middleware/demo bug: the middleware
heals free-form args correctly in the OSS-158 toolkit gate and against real
Gemini. Keep the test running (not skipped) as a deliberate demonstration for
the Showcase team; it does not red CI (these specs aren't run for google-adk in
CI). Update the spec header, the heal-test note, the fixture _note, the qa doc,
and the recovery_agent docstring to reflect this accurately (and drop the stale
sequenceIndex description). Tracked in Linear: aimock inner-subagent
disambiguation + recovery-demo LP parity.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The heal fixture relied on per-attempt aimock response switching (sequenceIndex
0=invalid -> 1=valid), but sequenceIndex doesn't advance across the recovery
loop's intra-turn render_a2ui calls in the per-integration Playwright context
(no X-Test-Id), so aimock served invalid on every attempt and the demo never
healed (0 metrics).
Switch the heal pill to a single deterministic response: the inner render_a2ui
returns FREE-FORM args (components/data as JSON strings) that the middleware
heals via parse_and_fix into a valid surface in one pass — no per-attempt
switching. Verified in-sandbox: free-form args -> a2ui_operations with 2 Metric
components. The exhaust pill is unchanged (structurally invalid every attempt ->
a2ui_recovery_exhausted hard-fail, already verified e2e).
Updates the agent docstring, qa checklist, and spec test name/comment to reflect
heal = parse_and_fix healing (not invalid->valid retry). aimock-fixtures schema
test: 818 passed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The clickPill helper (copied from declarative-gen-ui.spec.ts) matched the user
message with the stale '[data-message-role="user"]' selector, which no longer
exists in @copilotkit/react-core/v2 >= 1.60 — the user bubble is now
'[data-testid="copilot-user-message"]' (CopilotChatUserMessage). The message
dispatched fine; the helper just never found the bubble, timing out before the
heal/exhaust assertions ran. The sibling google-adk specs carry the same stale
selector (they don't run in CI for this integration, so it drifted unnoticed).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adding a2ui-recovery to the shared feature-registry cross-joins it across all
20 integrations (wired for google-adk, unshipped elsewhere), so the catalog
generator's hardcoded counts move: cross-join 920->940, metadata.total_cells
900->920, LGP 46->47 (unshipped 7->8). docs_only stays 20; the relative
sum-checks are unchanged. Fixes the Validate Showcase failure on #5662.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Playwright spec for the ADK-only a2ui-recovery demo, mirroring the AG-UI dojo
reference (apps/dojo/e2e/tests/adkMiddlewareTests/a2uiRecovery.spec.ts):
assert the STABLE end-states and skip the transient 'Retrying...' label
(threshold-gated/timing-dependent).
- heal pill: recovered surface paints (>=2 declarative-metric tiles), no
hard-failure UI, no render-error banners
- exhaust pill: 'Couldn't generate the UI' shows (A2UIRecoveryStates on the
a2ui_recovery_exhausted envelope), no faulty surface ever paints, chat stays usable
Selectors verified against @copilotkit/react-core/v2 A2UIRecoveryStates (failure
text) and the showcase declarative catalog testids. Clears the validate-parity
'no e2e spec' warning. Runs in the browser-e2e stack (frontend + aimock + agent_server).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Implements the Declarative Generative UI (A2UI dynamic schema) demo for both
the strands (Python) and strands-typescript integrations, bringing them to 1:1
parity with the canonical langgraph-python demo.
Backend:
- Add a dedicated dynamic-A2UI agent (a2ui_dynamic.py / buildA2uiDynamicAgent)
that wires no generate_a2ui tool. The runtime route sets
injectA2UITool: true + defaultCatalogId "declarative-gen-ui-catalog"; the
Strands adapter auto-injects generate_a2ui and drives a secondary render_a2ui
planner. StrandsAgentConfig.a2ui supplies the catalog id and a
composition_guide (sales dataset + composition rules) so the planner is
self-contained.
- Mount the agent on the /declarative-gen-ui sub-path and point the route's
HttpAgent at it, mirroring the a2ui-fixed-schema pattern.
- Bump ag_ui_strands 0.1.9 -> 0.2.1 (Python): the A2UI auto-injection config
(default_catalog_id, guidelines, render planner) landed in 0.2.0. Pulls
ag-ui-a2ui-toolkit transitively. strands-typescript already on 0.2.2.
Frontend:
- Port the langgraph-python declarative-gen-ui frontend (suggestions,
sales-context, definitions incl DataTable, renderers, chat) so the demo
matches the shared D5/D6 probe's sales-analyst pills and catalog components.
Fixtures:
- Author gen-ui-declarative D6 fixtures for both slugs from the langgraph-python
template (3 legs per pill: outer generate_a2ui, inner render_a2ui, narration).
Verified on real OpenAI (gpt-4o): both integrations paint a real A2UI surface
(4 Metric tiles + Revenue-by-Region PieChart + Monthly Revenue BarChart +
narration), matching the probe's expected testids and counts.
Note: the gen-ui-declarative D6 cell currently cannot go green because the
shared harness hits "ReferenceError: __name is not defined" in the probe's
page.evaluate (turn 1 preFill), before any agent call. That harness-tooling bug
is owned separately; the fixtures and wiring here are correct and the real-LLM
paint confirms the expected surface.
Strands mints UUID tool_call_ids for tool results (confirmed via the aimock
journal), so the langgraph-python toolCallId-keyed follow-up fixtures never
matched and the agent re-emitted the tool, looping (text-unstable) across
frontend-tools, gen-ui-agent, gen-ui-open, gen-ui-open-advanced,
gen-ui-headless-complete, reasoning-chain and the weather pills. Re-key the
affected multi-leg fixtures to the id- and thread-history-invariant
sequenceIndex pattern (as built-in-agent does), for both integrations.
Also:
- Add write_document poem/email/quantum fixtures for shared-state-streaming
(was a stale single fixture that 404d).
- Strip content from reasoning-chain tool legs (content+toolCalls in one
fixture is undefined behavior); reasoning rides toolCalls alone.
- Narrow the over-broad d4 summarize catch-all to "Summarize the sales
pipeline" so it stops shadowing gen-ui-agent competitor pill; matches
langgraph-python.
- Commit the real multimodal sample.png/sample.pdf (were git-LFS pointers
the harness could not resolve).
Add a write_document tool + document state hook to both Strands agents
so the shared-state-streaming demo can stream a document into
state.document (mirrors langgraph-python StateStreamingMiddleware). Strands
updates state from the full tool args, which the D6 probe tolerates.
Remove the get_weather stopStreamingAfterResult / stop_streaming_after_result
guard from both agents so weather concludes with a natural follow-up turn,
matching the langgraph-python gold standard; the accompanying multi-turn
id-invariant weather fixtures prevent the aimock replay loop without it.
The D6 fleet worker drives each integration over its insecure Docker
origin (http://<slug>:10000), where crypto.randomUUID is undefined so
hand-rolled headless chats threw and never mounted (sse-missing). tsx/
esbuild also wraps named inner functions in __name(...) calls that leak
into page.evaluate and throw __name is not defined.
Add installBrowserContextShims (init-scripts.ts): a __name no-op helper
and a crypto.randomUUID secure-context polyfill, registered via
addInitScript at document_start of every D6 page; wired into the
d6-all-pills newPage goto path.
0.2.2 ships the adapter fixes for the two upstream defects this hunt surfaced:
empty tool-result content (render-tool demos -> OpenAI 400) and RUN_FINISHED
emitted before parallel tool calls drain (INCOMPLETE_STREAM). With the bump,
beautiful-chat, the chart demos, and the catch-all 'Chain tools' pill complete
cleanly. Verified on real gpt-4o.
Both integrations: a dedicated `a2ui_fixed_schema` backend agent exposes a
`display_flight` tool that returns the A2UI `a2ui_operations` envelope
(createSurface -> updateComponents -> updateDataModel) built from a fixed,
pre-authored flight layout targeting the page's `copilotkit://flight-fixed-catalog`
catalog. Mirrors the upstream ag-ui dojo a2ui_fixed_schema demo (ag-ui#2021)
adapted to the showcase's existing catalog/renderers.
- TS returns the envelope as an object (lands in a json content block); Python
returns it as a JSON string (text block) — per each SDK's tool-return shape.
- Enable the runtime A2UIMiddleware on the a2ui-fixed-schema route
(injectA2UITool: false, defaultCatalogId pinned) so it detects the envelope
and paints; the agent emits the envelope itself, so no generate_a2ui injection.
- Mount the agent on the /a2ui-fixed-schema sub-path; point the route there.
Verified on real gpt-4o: the flight card paints on both integrations.
Add strands-typescript to BASELINE_PARTNERS so it gets its own coverage
column alongside its Python sibling (mirroring how langgraph-typescript sits
beside langgraph-python). Bump the partner-count assertion 26 -> 27.
Apply the same fixes to both the strands (Python) and strands-typescript
integrations for 1:1 parity:
- Lift RunAgentInput.context into the prompt (buildStatePrompt /
build_state_prompt) so useAgentContext values (readonly-state-agent-context)
and the openGenerativeUI design-skill / sandbox-function context actually
reach the model. The adapter does not surface context on its own; this
mirrors langgraph's lift-context-into-prompt pattern.
- open-gen-ui: prepend an imperative to the visualization design skill (and add
a design skill to the advanced cell) so the model calls generateSandboxedUi
instead of answering in plain text, and clarify that sandbox functions are
iframe->host bridges, not LLM tools.
- hitl-in-chat: sharpen the book_call description so it wins scheduling intents
over the shared backend schedule_meeting tool (which renders no picker).
- Add a roll_dice tool (shared python tools + TS tools) so the tool-rendering
catch-all 'Roll a d20' and 'Chain tools' pills work.
Add the missing "Show me my sales dashboard for this quarter." pill to
the pydantic-ai gen-ui-declarative D6 fixture set: an outer turn
(generate_a2ui, no args) plus the matching _design_a2ui_surface inner
turn carrying the dashboard component payload (KPI metrics row + revenue
pie + monthly-revenue bar), mirrored from the langgraph-python canonical
and the ms-agent-dotnet equivalent.
Closes the staging pydantic-ai D-chat 503 (no_fixture_match): the backend
hits aimock with tools=[generate_a2ui] for this userMessage under
x-aimock-context: pydantic-ai, but only ms-agent-dotnet had the fixture.
Deterministic canonical mirror (no real-LLM recording); contains only the
request-match shape and the A2UI response — no credentials.
The hand-rolled google.genai generate_a2ui planner (and the orphaned
SalesPipelineAgent that consumed it) in main.py are superseded by the
ag_ui_adk 0.7.0 middleware (get_a2ui_tool), now wired backend-owned in
declarative_gen_ui_agent.py / beautiful_chat_agent.py. main.py is reduced to
the shared tool wrappers + before_model/before_agent callbacks still covered
by tests; dead A2UI imports pruned.
- delete tests/python/test_generate_a2ui.py (tested the removed planner)
- manifest declarative-gen-ui: drop stale src/agents/main.py highlight + fix description
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The pool-fleet migration is complete: the control-plane harness plus the
prod workers (deployed 2026-06-19, HARNESS_ROLE=worker, pool count 2) now
cover every probe dimension the interim `harness-legacy` fleet-migration
bridge was holding live, so `harness-legacy` is dead config.
This is the code-side cleanup only:
- Remove the `harness-legacy` entry from the railway-envs SSOT and the
now-dead `key === "harness-legacy"` special-case in computePromoteClosure.
- Regenerate railway-envs.generated.json (41 -> 40 services).
- Drop harness-legacy from the golden snapshot, the gateIgnore expectation
sets, the promote-notify fixtures, and the redeploy-env doc comments;
update the service-count assertions (41 -> 40).
- Fix the stale "STAGING-ONLY" harness-workers comment: prod workers are
live on Railway, though this SSOT entry still models the staging
instance only (no prod env backfilled here yet).
The live Railway `harness-legacy` service is torn down separately as a
follow-up infra step.
Promote-notify Slack message: name the promoted AND failed services (one
Failed: header + bullets), legible "(N): <names>" count, real wall-clock
elapsed (integer-coerced), and drop the constant verify-prod legend line.
Durable healthcheckPath: track it per-service/env in the SSOT (railway-envs),
re-assert it on the promote pin path (omit-when-absent, never null), and route
deploy-to-railway provisioning through isTrackedService/resolveProvisionHealthcheck
so a tracked-null service omits the healthcheck while an untracked one keeps the
/api/health default — fixing the silent prod-healthcheck drift that refused aimock.
Tests: ruby pin-reassert spec + deploy-to-railway healthcheck spec + emit/golden/accessor.
Replace the hand-rolled google.genai A2UI planners (in main.py and
beautiful_chat_agent.py) with the published ag-ui-adk >= 0.7.0 middleware
sub-agent via get_a2ui_tool(), surfacing OSS-158 (forced render_a2ui
sub-agent + toolkit validate->retry recovery loop + recovery-exhausted
hard-fail envelope + render_as_llm_instructions / parse_and_fix healing).
Wiring is BACKEND-OWNED (injectA2UITool: false), matching the AWS Strands /
ag2 external-framework convention rather than langgraph-python's
runtime-driven injectA2UITool: true. Backend-owned is required: the planner
now lives in the ADK middleware, so letting the runtime also inject would
double-bind the tool slot. The explicit is load-bearing post
CopilotKit#5611 (a provider catalog otherwise defaults injectA2UITool to true).
- declarative_gen_ui_agent / beautiful_chat_agent: tools include
get_a2ui_tool({model, default_catalog_id}); beautiful keeps its other tools.
- shared_chat: add get_a2ui_model() to resolve a concrete Gemini BaseLlm for
the sub-agent (mirrors get_model's aimock-proxy wiring).
- routes: injectA2UITool stays false (declarative + beautiful-chat).
- registry/agent_server: no spec-level a2ui config needed (tool is agent-owned).
Verified vs published 0.7.0 + real Gemini: backend-wired generate_a2ui emits
a2ui_operations; OSS-158 gate (subagent + recovery invalid->valid + hard-fail
a2ui_recovery_exhausted) all retained through this exact wiring.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Pins the published middleware that ships the A2UI auto-inject + toolkit
recovery + render_as_llm_instructions/parse_and_fix path. Foundational
step; the A2UI agent re-wire onto that path follows under verification
once 0.7.0 is on PyPI.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
## Summary
- Add a dedicated `shell-docs-primary-cta` style for the hero quickstart
link so it keeps the primary CTA color inside reference content.
- Cover the new class usage in the hero and framework overview tests.
## Testing
- Updated unit tests to verify the quickstart CTA class and the matching
global CSS override.
- Updated unit tests to verify the framework overview markup includes
the primary CTA class.
The ag2 declarative-gen-ui route pointed its HttpAgent at the root
catch-all mount (agents/agent.py) instead of the dedicated
/declarative-gen-ui mount (a2ui_dynamic.py), and generate_a2ui declared
a required context arg that the model emits as {}. pydantic rejected
every call with "context Field required" and AG2 retried without bound —
a 630-iteration hot loop per pill that flooded logs and starved the
frontend.
Fix: route to the dedicated mount with injectA2UITool:false (the
dedicated agent owns generate_a2ui and emits a2ui_operations itself);
make generate_a2ui a no-arg tool matching the D6 fixtures and the
langgraph-python gold standard, with a constant inner system prompt
(per-pill distinctness comes from the captured user message). Regenerated
the gen-ui-declarative fixture and ported the LP definitions/renderers
catalog (all 7 driver testids) for parity. Eliminates the validation
loop: runsFinished=1, zero validation errors.
The LlamaIndex AG-UI adapter never forwarded RunAgentInput.tools, so
page-injected frontend tools (useFrontendTool / useComponent /
useHumanInTheLoop) were invisible to the LLM — runs the model couldn't
satisfy ended in RUN_ERROR (no RUN_FINISHED), which the harness reported
as sse-missing.
New RequestAwareAGUIChatWorkflow re-implements the chat step as the
upstream body plus three additions: forward RunAgentInput.tools as no-op
FunctionTool stubs carrying the verbatim injected JSON schema; re-role
tool-result messages to role="tool" on the LLM-bound message copy only
(so the hasToolResult second-leg fixture matches) without mutating
stored/snapshot history; and for frontend tool calls dispatch the
ToolCallEvent but skip the duplicate TOOL_CALL_CHUNK (the bare snapshot
already delivers the call). beautiful-chat 0/5 -> 5/5; mcp-apps fixed.
## Summary
The 12 `starter-*` services were deliberately `sleepApplication=true`
(sleepable), `probe.staging=false` (held out of the verify-deploy
staging matrix). That class-difference is the root of three recurring
symptoms: the `service=all` promote false-fails them (SLEEPING / no
running-digest), the deployed-starter smoke test 404s a cold container,
and they're excluded from staging validation. Decision: **bring them
into the normal managed-fleet flow — always-on + staging-probed.**
- **`verify-deploy.drivers.starter.ts` (new)** — real `probeStarter`
baseline driver (deployment SUCCESS + HTTP 200 on `/`, mirroring
`probeShell`); replaces the fail-loud `case "starter"` stub.
- **`railway-envs.ts`** — `staging.probe: false→true` for all 12
starters (prod already true); stale "staging probe OFF" comments
rewritten. `railway-envs.generated.json` regenerated (in-sync via
`--check`).
- **`provision-starter-fleet.ts`** — `sleepApplication: true→false` +
comments/logs; test updated.
## ⚠️ Deploy ordering (must hold)
This PR flips `probe.staging=true`, which routes starters into the
staging matrix. It must **not** merge until the live services are
flipped always-on, or a `service=all` promote would probe still-sleeping
starters and fail. Sequence: **(1) live-flip 24 instances
`sleepApplication=false` + redeploy (12 staging + 12 prod), (2) verify
awake + serving `/`, (3) merge this PR.**
## Verification
- Red-green: starter driver (stub→probe `/`), SSOT golden (probe.staging
true), provisioner (sleep false) — all RED→GREEN.
- `showcase/scripts`: 2092 tests pass; `tsc --noEmit` clean on changed
files.
## Test plan
- [x] vitest suites green (2092)
- [ ] CR
- [ ] live 24-instance always-on flip + redeploy (staging + prod)
- [ ] CI green
- [ ] post-flip: `service=all` dry-run shows starters pass staging probe
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Cover the data-copilot-running turn-done signal in waitForTurnComplete:
true->false transition completion, stayed-stopped quiescence, the
attr-gated early backstop, pre-send run-start baseline, and the
integration wait-for-turn-complete behavior.
Make waitForTurnComplete use the page-side data-copilot-running attribute
as the primary turn-done signal: detect the running true->false transition,
require stayed-stopped quiescence, and gate the early backstop on
attrPresent + runningNow to avoid headless false-RED. Capture a pre-send
run-start baseline so fast turns keep the primary signal alive, read
surfaceReady once per poll, and add computeMaxTurnDurationMs.
Add buildCopilotRunningObserverScript to sse-interceptor.ts and wire it
via addInitScript so the page exposes a data-copilot-running attribute
that the harness can observe for turn-completion signaling.
The built-in-agent shared-state-read-write demo wrote agent state via
direct property assignment (`agent.state = {...}`), which sets the value
but does not fire `onStateChanged` — so subscribed components never
re-render off a UI write. Every other framework demo (including the
langgraph-python gold standard) already uses `agent.setState(...)`, which
both updates state and notifies subscribers. This was the lone divergence.
Switch the seed and the title write to `agent.setState(...)` so the demo
matches the documented pattern (docs/shared-state.mdx says writes go
through `setState`) and the rest of the showcase.
Verified: built-in-agent D5 e2e-deep suite green after rebuild.
Flip the 12 starter-* services to behave like every other managed
showcase service: always-on and staging-probed.
- railway-envs SSOT: staging probe false -> true for all 12 starters
(prod was already on); refresh the stale "staging probe OFF" /
fail-loud-placeholder doc-comments to reflect baseline-driver verification.
- provision-starter-fleet: sleepApplication true -> false so the
provisioner creates always-on (not sleepable) Railway instances; update
the "the whole point — sleepable" header/inline comments + log lines.
- Regenerate railway-envs.generated.json (emit-railway-envs-json.ts) and
the golden fixture; update the SSOT probe + provisioner sleep assertions.
Replace the fail-loud `case "starter"` stub in verify-deploy's dispatch
switch with a real baseline liveness driver. The starter-template fleet
EXPOSEs only its Next.js frontend (serving `/` and `/api/copilotkit`,
NO `/api/health`), so the driver healthchecks `/` via probeBaseline,
exactly like the Next.js shells: deployment-SUCCESS + HTTP 200 on `/`.
Ran Shem Tov