Fix cua-ai ESM packaging, error-handling docs, API consistency - #18
Merged
Conversation
Switch packages/ai to NodeNext module resolution and add explicit .js extensions to relative imports so the published dist loads in plain Node ESM (previously failed with ERR_UNSUPPORTED_DIR_IMPORT). Run the full unit suite in CI and release instead of hard-coded file lists by excluding integration/live tests from the default vitest config, and add a post-pack ESM import smoke test to CI and the release workflow. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Family annotations now cover only the root id plus numeric revision or dated-snapshot suffixes, so named sibling variants like gpt-5.4-mini no longer list as CUA-capable. Drop gemini-2.5-computer-use-preview-10-2025 from the catalog: it rejects the function-declaration tools this package sends. Accept "gemini:" refs as an alias for the canonical "google:" prefix, name the valid providers in the unsupported-provider error, and add JSDoc to the models and api-keys entry points. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Standardize every provider namespace on <PROVIDER>_CUA_ACTION_TYPES and <PROVIDER>_COMPUTER_INSTRUCTIONS, restore anthropic's action-type export, and enforce the convention in tests. Align CUA_BATCH_TOOL_NAME with the computer_batch tool Anthropic actually ships and export ANTHROPIC_BATCH_TOOL_NAME. Make yutori.computerTools honor and validate its options instead of silently ignoring them, and document that the definitions are local mirrors stripped from the wire payload. Export and align the Yutori/Tzafon stream option interfaces (both now carry keepToolNames; redundant temperature/maxOutputTokens dropped), type mouse buttons as closed unions, export registerCuaProviders and make it re-register after pi-ai registry mutators clobber it, and thread tool options through resolveCuaRuntimeSpec. Harden the Yutori stream so one malformed tool call degrades to empty args instead of erroring the whole turn, restore api-literal typing on the Tzafon stream, and add unit coverage for Tzafon action normalization and argument unwrapping plus mocked-stream tests for both custom providers. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Document API key prerequisites and helpers, stopReason error handling, a multi-turn tool-result loop, the complete export surface, and per-provider canonical action subsets. Align the family-matcher docs with the tightened semantics and drop the removed Gemini 2.5 CUA preview model. Make the shipped quickstart import the package name (resolvable from the npm tarball), resolve its API key loudly, check stopReason, and switch providers via CUA_MODEL. Add packages/ai and packages/agent to the root README. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Add a 0.2.0 changelog entry covering the unreleased breaking changes since the 0.1.0 publish plus this release's fixes. Ship docs/ in the tarball and run the in-repo example through the source export condition. Update the release skill to full-suite test invocations and add a post-publish step that installs and imports the published package. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
rgarcia
marked this pull request as ready for review
June 10, 2026 18:26
|
Firetiger deploy monitoring skipped This PR didn't match the auto-monitor filter configured on your GitHub connection:
Reason: PR targets To monitor this PR anyway, reply with |
Replace the NodeNext + .js-specifier approach with a tsdown bundle: dist/ is a single ESM file with bundled-in relative modules, deps stay external, and src/test go back to extensionless imports. tsc -b remains for typechecking only, emitting declarations to a gitignored dist-tsc/ so project references keep working. Root build now builds cua-ai before tsc -b since dependents resolve its types from dist/. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Lead the loop section with the cua-agent pointer, drop the executor/ Yutori wire-format internals, the pi-ai versioning and registry side-effect notes, and the google-vs-gemini naming aside. Mark the legacy cua-* packages deprecated in the root README and make the "start here" pointer lead with cua-agent. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
tsdown evaluates its TypeScript config via Node's native type stripping (node >=22.18); on Node 20 it falls back to the optional unrun loader, which is not installed, so every job failed at the cua-ai build step. Node 22 also matches the >=22.19.0 engines floor the de-vendor PR introduces. Release workflows already run Node 24. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The exact 0.1.0 pin plus the workspace version bump made npm ci nest the published 0.1.0 (stale API, ESM-broken) under packages/agent, which broke the agent-e2e job at the build step. Pinning 0.2.0 resolves the workspace package again. The de-vendor PR sets the same value, so the branches still merge cleanly. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
Bugbot Autofix prepared fixes for both issues found in the latest run.
- ✅ Fixed: Node 20 engines mismatch
- Updated the root Node engine range to
^22.18.0 || >=24.0.0so advertised support matches the new tsdown build requirement.
- Updated the root Node engine range to
- ✅ Fixed: ESM smoke omits rejection handler
- Added explicit
.catch(...)handlers to both ESM smoke-test import checks so import failures always exit non-zero.
- Added explicit
Or push these changes by commenting:
@cursor push 6260abd7ef
Preview (6260abd7ef)
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -31,7 +31,7 @@
cd "$RUNNER_TEMP/esm-smoke"
npm init -y
npm install "$RUNNER_TEMP"/onkernel-cua-ai-*.tgz
- node --input-type=module -e "import('@onkernel/cua-ai').then((m) => { if (typeof m.getCuaModel !== 'function') process.exit(1); })"
+ node --input-type=module -e "import('@onkernel/cua-ai').then((m) => { if (typeof m.getCuaModel !== 'function') process.exit(1); }).catch((err) => { console.error(err); process.exit(1); })"
integration:
runs-on: ubuntu-latest
diff --git a/.github/workflows/release-cua-ai.yml b/.github/workflows/release-cua-ai.yml
--- a/.github/workflows/release-cua-ai.yml
+++ b/.github/workflows/release-cua-ai.yml
@@ -71,7 +71,7 @@
cd "$RUNNER_TEMP/esm-smoke"
npm init -y
npm install "$RUNNER_TEMP"/onkernel-cua-ai-*.tgz
- node --input-type=module -e "import('@onkernel/cua-ai').then((m) => { if (typeof m.getCuaModel !== 'function') process.exit(1); })"
+ node --input-type=module -e "import('@onkernel/cua-ai').then((m) => { if (typeof m.getCuaModel !== 'function') process.exit(1); }).catch((err) => { console.error(err); process.exit(1); })"
- name: Publish to npm
run: npm publish --workspace @onkernel/cua-ai --access public
diff --git a/package.json b/package.json
--- a/package.json
+++ b/package.json
@@ -24,7 +24,7 @@
"clean": "tsc -b --clean && npm run clean:native --workspace @onkernel/ptywright --if-present"
},
"engines": {
- "node": ">=20"
+ "node": "^22.18.0 || >=24.0.0"
},
"devDependencies": {
"@types/node": "22.18.4",You can send follow-ups to the cloud agent here.
Reviewed by Cursor Bugbot for commit 7880147. Configure here.
| "build:cli": "npm run build --workspace @onkernel/cua-cli", | ||
| "dev": "tsc -b --watch", | ||
| "typecheck": "tsc -b", | ||
| "typecheck": "npm run build --workspace @onkernel/cua-ai && tsc -b", |
There was a problem hiding this comment.
Node 20 engines mismatch
Medium Severity
The root engines.node field still allows Node 20, but @onkernel/cua-ai now builds with tsdown, which requires Node ^22.18.0 or ≥24. CI was bumped to 22 while the repo still advertises ≥20, so a normal npm run build or root typecheck can fail on Node 20 even though the project claims support.
Reviewed by Cursor Bugbot for commit 7880147. Configure here.
| cd "$RUNNER_TEMP/esm-smoke" | ||
| npm init -y | ||
| npm install "$RUNNER_TEMP"/onkernel-cua-ai-*.tgz | ||
| node --input-type=module -e "import('@onkernel/cua-ai').then((m) => { if (typeof m.getCuaModel !== 'function') process.exit(1); })" |
There was a problem hiding this comment.
ESM smoke omits rejection handler
Low Severity
The new post-pack ESM smoke step only chains .then() on import('@onkernel/cua-ai') and never handles rejection. If the tarball fails to load, the check may not exit with a non-zero code reliably, so CI could pass without proving the artifact is importable.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit 7880147. Configure here.
This comment has been minimized.
This comment has been minimized.
7 tasks
|
Bugbot Autofix prepared a fix for the issue found in the latest run.
Or push these changes by commenting: Preview (ebccaacc97)diff --git a/packages/ai/src/providers/common.ts b/packages/ai/src/providers/common.ts
--- a/packages/ai/src/providers/common.ts
+++ b/packages/ai/src/providers/common.ts
@@ -442,6 +442,8 @@
export interface CuaPayloadContext {
/** Tool names that should remain in the outbound provider payload even if the provider strips local CUA executors. */
keepToolNames?: readonly string[];
+ /** Canonical CUA actions enabled for this runtime, if narrowed by options. */
+ actions?: readonly CuaActionType[];
}
export type CuaPayloadHook = (payload: unknown, model: Model<Api>, context?: CuaPayloadContext) => unknown | Promise<unknown>;
diff --git a/packages/ai/src/providers/tzafon/provider.ts b/packages/ai/src/providers/tzafon/provider.ts
--- a/packages/ai/src/providers/tzafon/provider.ts
+++ b/packages/ai/src/providers/tzafon/provider.ts
@@ -117,14 +117,20 @@
if (!payload || typeof payload !== "object") return undefined;
const current = payload as { tools?: unknown };
const keepToolNames = new Set(context?.keepToolNames ?? []);
+ const useNativeComputerUse = shouldUseTzafonNativeComputerUse(context);
+ const allowedActions = new Set(context?.actions ?? []);
const existingTools = Array.isArray(current.tools) ? current.tools : [];
- const shouldAddComputerUse = existingTools.some((tool) => {
- const name = readToolName(tool);
- return Boolean(name && TZAFON_LOCAL_ACTION_TOOL_NAMES.has(name) && !keepToolNames.has(name));
- });
+ const shouldAddComputerUse = useNativeComputerUse
+ ? existingTools.some((tool) => {
+ const name = readToolName(tool);
+ return Boolean(name && TZAFON_LOCAL_ACTION_TOOL_NAMES.has(name) && !keepToolNames.has(name));
+ })
+ : false;
const tools = existingTools.filter((tool) => {
const name = readToolName(tool);
- return !name || keepToolNames.has(name) || !TZAFON_LOCAL_ACTION_TOOL_NAMES.has(name);
+ if (!name || keepToolNames.has(name) || !TZAFON_LOCAL_ACTION_TOOL_NAMES.has(name)) return true;
+ if (!useNativeComputerUse && allowedActions.size > 0) return allowedActions.has(name as (typeof CUA_ACTION_TYPES)[number]);
+ return !useNativeComputerUse;
});
return {
...(payload as Record<string, unknown>),
@@ -132,6 +138,12 @@
};
}
+function shouldUseTzafonNativeComputerUse(context?: CuaPayloadContext): boolean {
+ if (!context?.actions) return true;
+ const allowed = new Set(context.actions);
+ return CUA_ACTION_TYPES.every((action) => allowed.has(action));
+}
+
/** Derive a unique canonical tool-call id for a Tzafon computer action. */
export function tzafonToolCallId(callId: string, actionIndex: number): string {
return actionIndex === 0 ? callId : `${callId}:${actionIndex}`;
diff --git a/packages/ai/src/providers/yutori/provider.ts b/packages/ai/src/providers/yutori/provider.ts
--- a/packages/ai/src/providers/yutori/provider.ts
+++ b/packages/ai/src/providers/yutori/provider.ts
@@ -17,6 +17,7 @@
isYutoriLocalActionToolName,
toCanonicalActions,
yutoriToolSetForModel,
+ YUTORI_CUA_ACTION_TYPES,
YUTORI_N15_EXPANDED_ACTION_TYPES,
} from "./actions";
import { canonicalToolCallArguments, canonicalToolCallName, type CuaPayloadContext } from "../common";
@@ -45,20 +46,31 @@
if (!payload || typeof payload !== "object") return undefined;
const current = payload as { tools?: unknown };
const keepToolNames = new Set(context?.keepToolNames ?? []);
+ const useNativeToolSet = shouldUseYutoriNativeToolSet(context);
+ const allowedActions = new Set<string>(context?.actions ?? []);
const tools = Array.isArray(current.tools)
? current.tools.filter((tool) => {
const name = readToolName(tool);
- return !name || keepToolNames.has(name) || !isYutoriLocalActionToolName(name);
+ if (!name || keepToolNames.has(name) || !isYutoriLocalActionToolName(name)) return true;
+ if (!useNativeToolSet && allowedActions.size > 0) return allowedActions.has(name);
+ return !useNativeToolSet;
})
: undefined;
- const toolSet = model ? yutoriToolSetForModel(model.id) : undefined;
+ const toolSet = useNativeToolSet && model ? yutoriToolSetForModel(model.id) : undefined;
return {
...(payload as Record<string, unknown>),
+ ...(useNativeToolSet ? {} : { tool_set: undefined, disable_tools: undefined }),
...(toolSet ? { tool_set: toolSet, disable_tools: [...YUTORI_N15_EXPANDED_ACTION_TYPES] } : {}),
...(tools && tools.length > 0 ? { tools } : { tools: undefined }),
};
}
+function shouldUseYutoriNativeToolSet(context?: CuaPayloadContext): boolean {
+ if (!context?.actions) return true;
+ const allowed = new Set(context.actions);
+ return YUTORI_CUA_ACTION_TYPES.every((action) => allowed.has(action));
+}
+
async function runYutoriStream(
stream: ReturnType<typeof createAssistantMessageEventStream>,
model: Model<Api>,
diff --git a/packages/ai/src/runtime-spec.ts b/packages/ai/src/runtime-spec.ts
--- a/packages/ai/src/runtime-spec.ts
+++ b/packages/ai/src/runtime-spec.ts
@@ -32,6 +32,13 @@
const model = typeof input === "string" ? getCuaModel(input) : input;
const provider = providerForModel(model);
const mod: CuaProviderModule = PROVIDERS[provider];
+ const onPayload = mod.onPayload
+ ? (payload: unknown, requestModel: typeof model, context?: Parameters<NonNullable<CuaProviderModule["onPayload"]>>[2]) =>
+ mod.onPayload?.(payload, requestModel, {
+ ...context,
+ ...(options?.actions ? { actions: options.actions } : {}),
+ })
+ : undefined;
return {
model,
provider,
@@ -40,6 +47,6 @@
defaultSystemPrompt: mod.buildSystemPrompt(),
coordinateSystem: mod.coordinateSystem(),
screenshot: mod.screenshot,
- onPayload: mod.onPayload,
+ onPayload,
};
}
diff --git a/packages/ai/test/runtime-spec.test.ts b/packages/ai/test/runtime-spec.test.ts
--- a/packages/ai/test/runtime-spec.test.ts
+++ b/packages/ai/test/runtime-spec.test.ts
@@ -53,5 +53,9 @@
const yutoriSpec = resolveCuaRuntimeSpec("yutori:n1.5-latest", { actions: ["click"] });
expect(yutoriSpec.toolDefinitions).toEqual([]);
expect(yutoriSpec.toolExecutors.map((executor) => executor.definition.name)).toEqual(["click"]);
+
+ const tzafonSpec = resolveCuaRuntimeSpec("tzafon:tzafon.northstar-cua-fast", { actions: ["click"] });
+ expect(tzafonSpec.toolDefinitions.map((tool) => tool.name)).toEqual(["click"]);
+ expect(tzafonSpec.toolExecutors.map((executor) => executor.definition.name)).toEqual(["click"]);
});
});
diff --git a/packages/ai/test/tzafon-payload.test.ts b/packages/ai/test/tzafon-payload.test.ts
--- a/packages/ai/test/tzafon-payload.test.ts
+++ b/packages/ai/test/tzafon-payload.test.ts
@@ -42,6 +42,25 @@
]);
});
+ it("keeps narrowed local action tools and skips native computer_use", () => {
+ const payload = {
+ tools: [
+ { type: "function", name: "click" },
+ { type: "function", name: "move" },
+ { type: "function", name: "custom_tool" },
+ ],
+ };
+
+ const next = tzafon.tzafonComputerUseOnPayload(payload, undefined, {
+ actions: ["click"],
+ }) as { tools?: Array<{ type?: string; name?: string }> };
+
+ expect(next.tools).toEqual([
+ { type: "function", name: "click" },
+ { type: "function", name: "custom_tool" },
+ ]);
+ });
+
it("returns undefined for non-object payloads", () => {
expect(tzafon.tzafonComputerUseOnPayload(undefined)).toBeUndefined();
expect(tzafon.tzafonComputerUseOnPayload("x")).toBeUndefined();
diff --git a/packages/ai/test/yutori-payload.test.ts b/packages/ai/test/yutori-payload.test.ts
--- a/packages/ai/test/yutori-payload.test.ts
+++ b/packages/ai/test/yutori-payload.test.ts
@@ -39,6 +39,26 @@
expect(next.tools?.map((tool) => tool.function?.name)).toEqual(["batch_computer_actions"]);
});
+ it("keeps narrowed local action tools and skips the native tool set", () => {
+ const payload = {
+ tools: [
+ { type: "function", function: { name: "click" } },
+ { type: "function", function: { name: "move" } },
+ { type: "function", function: { name: "custom_tool" } },
+ ],
+ };
+ const next = yutori.yutoriNativeToolSetOnPayload(payload, { id: "n1.5-latest" } as never, {
+ actions: ["click"],
+ }) as {
+ tool_set?: string;
+ disable_tools?: string[];
+ tools?: Array<{ function?: { name?: string } }>;
+ };
+ expect(next.tool_set).toBeUndefined();
+ expect(next.disable_tools).toBeUndefined();
+ expect(next.tools?.map((tool) => tool.function?.name)).toEqual(["click", "custom_tool"]);
+ });
+
it("returns undefined for non-object payloads", () => {
expect(yutori.yutoriNativeToolSetOnPayload(undefined)).toBeUndefined();
expect(yutori.yutoriNativeToolSetOnPayload("x")).toBeUndefined();You can send follow-ups to the cloud agent here. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
packages/aidist/is now bundled with tsdown (single ESM file, deps external, source keeps extensionless imports;tsc -bstays for typechecking via a gitignoreddist-tsc/), so the published tarball imports under plain Node ESM (0.1.0 fails withERR_UNSUPPORTED_DIR_IMPORT). CI andrelease-cua-ai.ymlgain a pack + install-into-temp-dir + ESM import smoke step that would have caught the 0.1.0 regression.complete()/stream()resolve withstopReason: "error"instead of throwing; the README Quick Start, a new Error Handling section, andexamples/quickstart.tsnow checkstopReason, so a bad API key fails loudly instead of exiting 0 with no output. A new Continuing the Loop section documents theToolResultMessageshape and a two-turn example.api-keyshelpers documented, and the quickstart resolves its key viarequireCuaEnvApiKeyForModel. The shipped example now imports@onkernel/cua-ai(resolvable from the tarball), switches providers viaCUA_MODEL, anddocs/ships in the npm files.<PROVIDER>_CUA_ACTION_TYPES,<PROVIDER>_COMPUTER_INSTRUCTIONS,computerTools/computerToolExecutors/createActionSchema/coordinateSystem/providerModule,<Provider>Action,ComputerToolsOptions), enforced by a parity test.CUA_BATCH_TOOL_NAMEis now"computer_batch"(matching the only shipped batch tool) withanthropic.ANTHROPIC_BATCH_TOOL_NAMEexported.yutori.computerToolshonors and validates{ actions }.registerCuaProviders()is exported and re-registers after pi-ai registry mutators.resolveCuaRuntimeSpec(input, options?)threads tool options through. Mouse buttons are typed as closed unions (CuaMouseButton/CuaDragMouseButton; wire schemas unchanged). Family annotations match only root + numeric revision/dated-snapshot suffixes, sogpt-5.4-mini/-nano/-prono longer list as CUA-capable.gemini-2.5-computer-use-preview-10-2025is removed from the catalog (rejects the function declarations this package sends);gemini:refs are accepted as an alias forgoogle:.YutoriOptions/TzafonResponsesOptionsare exported and aligned onkeepToolNames; malformed Yutori tool calls degrade per-call instead of erroring the turn. JSDoc added acrossmodels.tsandapi-keys.ts.cua-*packages deprecated — cua-ai/cua-agent are the canonical source — and the "start here" pointer leads with cua-agent.vitest.integration.config.ts+npm run test:integration;release-cua-ai.ymldrops its stale 6-file list..agents/skills/release/SKILL.mduses the plain full-suite invocations and adds a post-publish install + import verification step.Merge coupling: this PR also bumps
packages/agent's@onkernel/cua-aipin to0.2.0so the workspace package resolves locally (an exact0.1.0pin plus the 0.2.0 workspace version madenpm cinest the stale, ESM-broken published 0.1.0 underpackages/agent, breaking the agent-e2e build). The de-vendor PR (#17) sets the same line to the same value, so the branches merge cleanly. Publishing order still applies:cua-ai/v0.2.0beforecua-agent.Test plan
npm cifrom scratch, thennpm run build --workspace @onkernel/cua-ai(tsdown) — greendist/is a bundled ESM entry (no relative runtime imports beyond the emitted chunk)npm run test:integration --workspace @onkernel/cua-aicollects only the integration file; ran live with available keys (2 passed, 3 skipped)npm run typecheck(tsc -b, includes packages/agent via project reference) — greenCUA_BATCH_TOOL_NAME === "computer_batch", namespace parity, action-subset counts,gemini:alias)--conditions=source; both fail loudly with the missing-key error before any request; example typechecks standaloneStopReason/ToolResultMessageshapes, pi-ai google env fallback (GEMINI_API_KEYonly), per-provider action subsets (anthropic missing back/forward/url; yutori missing screenshot/url/cursor_position), env-var table order, button-coercion semanticscua-ai/v0.1.0tag (e.g.maxOutputTokenswas accepted in 0.1.0)🤖 Generated with Claude Code
Note
Medium Risk
0.2.0 introduces multiple documented breaking API and catalog changes; CI/release now gate on ESM smoke tests, but consumers and any code still on 0.1.0 naming must migrate.
Overview
@onkernel/cua-ai 0.2.0 fixes plain Node ESM consumption by bundling
dist/with tsdown (replacing extensionlesstscoutput that brokeimport "@onkernel/cua-ai"), shipsdocs/in the tarball, and bumps@onkernel/cua-agentto depend oncua-ai@0.2.0. Rootbuild/typechecknow buildcua-aifirst via tsdown.CI and release move to Node 22, run the full
cua-aiunit suite (integration viatest:integration), and add pack + ESM import smoke in CI andrelease-cua-ai.yml; the release skill documents full-suite tests and post-publish import checks.Public API and docs (many 0.2.0 breaking changes): unified provider namespace exports,
CUA_BATCH_TOOL_NAME→"computer_batch", stricter model family matching, removal ofgemini-2.5-computer-use-previewfrom the catalog,gemini:alias forgoogle:, exportedregisterCuaProviders()andresolveCuaRuntimeSpec(..., options), Yutori/Tzafon stream hardening, and expanded README (API keys,stopReason, tool loop). Root README positionscua-ai/cua-agentas canonical and marks legacycua-*packages deprecated.Reviewed by Cursor Bugbot for commit 7880147. Bugbot is set up for automated code reviews on this repo. Configure here.