Two silent failures in the Python SDK prevented frontend context from reaching the agent:
1. langgraph_default_merge_state rebuilt copilotkit as {actions} only, dropping the
useCopilotReadable items the TypeScript runtime had placed in state.copilotkit.context.
2. CopilotKitContext.properties were never forwarded through execute_agent → agent.execute(),
so <CopilotKit properties={...} /> values never reached state.copilotkit.properties.
Both fields now land cleanly under state["copilotkit"] on every agent invocation.
Also adds tests, a showcase example, and a new docs page.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
## Summary
- Remove the Open GenUI rendering strategy from all 17 showcase starters
- Clean up all supporting files: feature registries, constraints, e2e
tests, Python middleware, docs
- Regenerate all starters from updated template
### What was removed
The "open-genui" render mode allowed agents to generate arbitrary HTML
rendered in a sandboxed iframe. This is being extricated from starters
into a separate project. The core SDK `OpenGenerativeUIRenderer`
component is **not** affected — only the showcase starter UI option.
### Files changed
- **Template**: Deleted `open-genui/` renderer, updated types.ts,
page.tsx, suggestions hook
- **Registry**: Removed `open-gen-ui` and `byoc-opengenui` features from
all registries (shell, shell-dojolike, shared)
- **Constraints**: Removed "open" constraint block and open-gen-ui
references
- **E2E tests**: Removed Open GenUI test cases from starter-e2e,
screenshots, renderer-selector specs
- **Python middleware**: Removed OPEN_GENUI_INSTRUCTION and open-genui
handling from render_mode.py
- **Starters**: Regenerated all 17 via generate-starters.ts
- **Docs**: Updated QA-COVERAGE.md pill count from 5 to 4
### Local verification
- Docker build of langgraph-python starter: success
- Rendered page shows exactly 4 render modes (Tool-Based, A2UI,
json-render, HashBrown)
- Zero "open-genui" references in rendered HTML
- 521 validation tests pass
## Test plan
- [x] Docker build succeeds locally
- [x] Rendered starter shows 4 modes, no Open GenUI
- [x] Zero grep matches for open-genui across showcase/
- [x] 521 vitest validation tests pass
- [ ] CI green
🤖 Generated with [Claude Code](https://claude.com/claude-code)
## Summary
Fixes the last 2 crashing starters (claude-sdk-typescript + mastra)
after PR #3933 resolved 5/7.
1. **ESM resolution failure** (claude-sdk-typescript) — nested
`agent/package.json` in Docker runner stage creates a separate package
boundary that confuses tsx module resolution. Fix: `rm -f
agent/package.json` in Dockerfile template after COPY.
2. **Recursive chown** (all templates) — `chown app:app /app` only
changed the directory, not contents. Mastra needs to create
`/app/src/mastra/public` at runtime. Fix: `chown -R app:app /app` in all
Dockerfile templates.
3. **Mastra version drift** (mastra) — floating `beta` dist-tags resolve
to incompatible versions between `mastra` CLI and `@mastra/core`. Fix:
generation script pins mastra ecosystem deps to stable versions via
`PIN_OVERRIDES` map.
All fixes verified locally with Docker builds before pushing.
## Test plan
- [x] Docker build + run claude-sdk-typescript locally — agent starts on
port 8123
- [x] Docker build + run mastra locally — `mastra 1.6.0 ready in 675 ms`
- [x] 502 generate-starters tests pass
- [x] Drift check passes
- [ ] Deploy to Railway and verify all 17 starters return 200
## Summary
- The docs-sync warning notification was sending "see workflow run for
details" with no actionable information
- Now reads `review-items.txt` and includes the file list directly in
the Slack message
- Recipients can see which files need attention without digging through
CI logs
## Test plan
- [ ] Trigger docs-sync with a file that has showcase-local
modifications (exit code 3 path)
- [ ] Verify Slack notification includes the file list in a code block
- [ ] Verify auto-push-only path (exit code 0) still does NOT send the
warning notification
The source package's src/agent/ was just committed — regenerating
the starter now correctly copies agent/index.ts and agent/package.json
into the starter directory, fixing the drift check.
The src/agent/ directory was never committed to the repo, causing
the drift-check to fail in CI. The starter was generated from a
local-only copy of this directory.
The warning notification for files needing manual review was sending
'see workflow run for details' with no actionable information.
- Read review-items.txt and include file list in the Slack message
- Use jq for proper JSON escaping (handles newlines, quotes, special chars)
- Guard against missing review-items.txt with fallback and ::warning::
- Review-needed notification fires independently of push/merge outcome
## Summary
- Add deployed starter L1-L3 smoke tests to `integration-smoke.spec.ts`
(health, agent endpoint, chat round-trip) hitting live Railway
`demo_url`s from `registry.json`
- Upgrade existing L4 `@tools` test with DOM element assertion
(`Promise.all` + `.some()`) to catch bugs like the `query_data` agent
loop that produced text but never rendered charts
- Add `deployed` boolean field to all 17 starter objects in
`registry.json` (14 healthy, 3 currently down)
- New `starter_deployed_smoke.yml` CI workflow: 6h cron + post-deploy
trigger + manual dispatch, with Slack/GitHub issue alerting on failure
### Context
7+ starters were crashing on Railway for hours with no automated
detection. The existing test suite only covered integration backends —
starters had zero deployed-URL testing. The
`agent-looping-on-query_data` bug shipped undetected because L4 only
checked text keywords, not rendered components.
### Key design decisions
- URLs read dynamically from `registry.json` (not hardcoded) — new
starters auto-tested
- `deployed` field gates which starters run (opt-in via `=== true`),
with `SMOKE_ALL` override
- Single-test-per-starter pattern avoids `fullyParallel` config
overriding serial mode
- Rendered-tools check is a warning (not hard failure) pending
data-testid standardization
- `STARTER_SLUG` env var supports single-starter manual dispatch
filtering
### Spec
https://www.notion.so/copilotkit/3433aa3818528112b2b6f65a4fab92d4
## Test plan
- [ ] Verify `@starter-health` tests pass against the 14 deployed
starters
- [ ] Verify `@starter-agent` tests pass against deployed starters
- [ ] Verify `@starter-chat` tests produce assistant responses on
deployed starters
- [ ] Verify 3 non-deployed starters (mastra, crewai-crews,
claude-sdk-typescript) are skipped by default
- [ ] Verify `SMOKE_ALL=true` includes non-deployed starters
- [ ] Verify `STARTER_SLUG=langgraph-python` filters to a single starter
- [ ] Verify rendered-tools warning fires for integrations without chart
components
- [ ] Verify CI workflow triggers correctly on manual dispatch
- [ ] Verify Slack + GitHub issue creation on failure path
🤖 Generated with [Claude Code](https://claude.com/claude-code)
1. Remove nested agent/package.json in TypeScript Dockerfile template to
prevent ESM module resolution failure (claude-sdk-typescript)
2. Use chown -R for /app in all Dockerfile templates so runtime-created
dirs (e.g. /app/src/mastra/public) are writable by non-root user
3. Pin mastra ecosystem deps to stable versions (1.6.0/1.25.0) in
generation script — floating beta tags caused version drift between
mastra CLI and @mastra/core
New workflow running starter health/agent/chat tests against Railway:
- Triggers: 6h cron, after showcase deploy, manual dispatch
- Alerts on schedule + workflow_run failures (Slack + GitHub issue)
- Issue dedup by title match, continue-on-error on Slack
- Proper working-directory for npm ci and Playwright install
Support retries with configurable delay in checkHealth() for Railway
cold-start scenarios. Captures last error from retry loop instead of
making a redundant final request.
Add a `deployed` boolean to all 17 starter sub-objects. 14 marked
true (healthy on Railway), 3 marked false (mastra, crewai-crews,
claude-sdk-typescript — currently down). Used by smoke tests to
gate which starters are tested.
## Release monorepo v1.56.0
**Scope:** `monorepo` | **Bump:** `minor`
---
### How this release process works
1. **This PR was created automatically** by the "release / create-pr"
workflow.
It bumped the `monorepo` packages to `1.56.0`
and generated AI-enhanced release notes.
2. **CI runs on this PR** — the full test suite (unit tests, lint, type
checks, build)
must pass before merging. This is the review gate.
3. **Review the release notes** in `release-notes.md` in this PR.
If a Notion draft was created, you can edit the release notes there
before merging.
4. **When this PR is merged**, the `release / publish` workflow
automatically:
- Builds all packages
- Publishes the `monorepo` packages to npm at version `1.56.0`
- Creates git tag `monorepo/v1.56.0`
- Creates a GitHub Release with the final release notes
### Before merging
- [ ] CI is green (tests, lint, types, build)
- [ ] Version bumps look correct
- [ ] Release notes are accurate (edit in Notion if a draft was created)
---
> **Do not merge until CI is fully green.** The full test suite runs
automatically on this PR.
## Summary
7 out of 17 Railway-deployed starters are returning 502 (CRASHED). Three
distinct root causes:
1. **EACCES `/home/app`** (langgraph-typescript, claude-sdk-typescript,
mastra) — `npx` at runtime needs a home dir for its cache, but the
non-root `app` user was created with `--no-create-home`. Fix: `mkdir -p
/home/app && chown app:app /home/app` in all 4 Dockerfile templates.
2. **IndexError `Path.parents[4]`** (agno, claude-sdk-python,
ms-agent-python) — `search_flights.py` schema fallback accesses
`parents[4]` but Docker path `/app/agent/tools/search_flights.py` only
has 4 parents (0-3). Fix: wrap in try/except IndexError in shared source
+ regenerate.
3. **Wrong import path** (langroid, crewai-crews) — `from agents.X
import` but starters use `agent/` (singular). `rewritePythonImports()`
in generate-starters.ts now converts these to relative imports.
All fixes applied to upstream sources (templates, shared python,
generation script) so drift check passes.
## Test plan
- [x] 251 generate-starters tests pass
- [x] Drift check passes (`--check`)
- [x] Pre-commit hooks pass (lint, format, unit tests)
- [ ] Deploy to Railway and verify all 17 starters return 200
1. EACCES on /home/app: Dockerfile templates now create home dir for
non-root user before USER app (langgraph-ts, claude-sdk-ts, mastra)
2. IndexError Path.parents[4]: search_flights.py schema fallback now
guards against shallow Docker paths (agno, claude-sdk-py, ms-agent-py)
3. Wrong import path: rewritePythonImports() now converts
from agents.X to relative imports in starter agent dirs (langroid, crewai)
## Summary
- Add `debug` option to `CopilotRuntime` constructor and `<CopilotKit>`
provider for detailed event pipeline logging
- Server-side uses Pino structured logger; client-side uses
`console.debug` at each AG-UI pipeline stage
- Accepts `true` for full output or a granular config: `{ events,
lifecycle, verbose }`
- The two toggles are independent — enabling debug on the client does
not affect the server
## Changes
**`@copilotkit/shared`**
- New `DebugConfig` type, `ResolvedDebugConfig` interface,
`resolveDebugConfig()` normalizer
**`@copilotkit/runtime`**
- `debug` option on `CopilotRuntime({ debug: true })`
- Lifecycle logging: agent run started, SSE stream
opened/completed/errored
- Event logging: every emitted event (verbose = full payload, otherwise
key fields summary)
**`@copilotkit/react-core` + `@copilotkit/core`**
- `debug` prop on `<CopilotKit debug={true}>`
- Threads config through `CopilotKitCore` → `AgentRegistry` →
`ProxiedCopilotRuntimeAgent`
**Companion PR:** ag-ui-protocol/ag-ui#1398 (AG-UI client-side debug
logging)
## Test plan
- [x] All 647 runtime tests pass
- [x] All 352 AG-UI client tests pass
- [x] Lint clean (0 errors)
- [x] All affected packages build successfully
- [ ] Manual: set `debug: true` on runtime and provider, verify
structured logs on server and console output on client
🤖 Generated with [Claude Code](https://claude.com/claude-code)
The docs table and prose claimed `debug: true` sets `verbose: true`,
but the implementation intentionally defaults verbose to false (PII
safety). Fixed the table and explanatory text to match.
Also removed packages/vscode-extension/README.md which was committed
on this branch by mistake — it describes an unrelated VS Code extension
and has nothing to do with debug mode.
## Summary
- Replace custom `{ name, props }` schema format with A2UI v0.9
spec-aligned inline catalog format (`allOf` + `properties`) — eliminates
LLM "props" nesting confusion
- Restructure generation prompts: inline literal values are the default,
path binding is a narrow schema-driven exception
- Bump `@ag-ui/a2ui-middleware` to `0.0.5` in runtime
## Context
LLMs generating dynamic A2UI sometimes produced invalid output:
1. Nesting properties under a `"props"` key (because our schema used
`props` as a key)
2. Using `{ "path": "..." }` bindings on properties that only accept
literal values (e.g. chart `data`, metric `value`)
Both issues caused silent render failures or runtime crashes.
## Test plan
- [x] Tested with ag-ui dojo — A2UI dynamic schema demo works
- [x] Tested with CopilotKit langgraph-starter — sales dashboard renders
correctly
- [x] `@ag-ui/a2ui-middleware@0.0.5` published to npm
🤖 Generated with [Claude Code](https://claude.com/claude-code)
The generate-starters script was flattening src/main/{java,resources} into
agent/{java,resources}, breaking Maven builds. Add a post-copy step that
moves them to agent/src/main/{java,resources} as Maven expects.
Replace TODO placeholders in all 73 QA checklist files across 17 showcase
packages with specific, actionable test steps derived from reading the
actual demo page source code. Each checklist now references real
data-testid selectors, hook names, expected UI states, and interaction
flows.
Add showcase/QA-COVERAGE.md with a comprehensive matrix showing coverage
across manual QA, Vitest unit tests, Playwright E2E (smoke and interaction),
per-package E2E, aimock fixtures, and CI automation for all 9 demos plus
the Sales Dashboard starter hero.
The agent's system prompt didn't mention pieChart/barChart frontend
tools, so gpt-4o-mini would call query_data repeatedly instead of
rendering charts. Updated the system prompt to explicitly instruct
the agent to call query_data once then use pieChart/barChart, and
updated the query_data tool description to reinforce single-call
semantics.
Martha Schumann