feat(agents): OpenCodeHarness session events, and Workspace tools in the example by mattzcarey · Pull Request #2568 · cloudflare/agents

mattzcarey · 2026-10-12T12:22:13Z

OpenCodeHarness's session.events() now returns a snapshot of the session and then its events, in the same shape as PiHarness and ContainerHarness. The OpenCode example gives the model the Workspace tools from @cloudflare/computer (read, write, edit, delete, ls, find, grep, and exec with git) in place of its notes plugin.

This takes the parts of #2483 (thanks @aron-cf) we want on top of the harness from #2513:

the event union, so one client can follow any harness instead of decoding OpenCode's ~90 raw event types
the Workspace tools, so OpenCode has files and code to work with in a Durable Object
harness.fetch for the OpenCode CLI is left out: OpenCode doesn't expose its in-process handler, and getting it means patching OpenCode.make. Tracked in OpenCodeHarness: serve OpenCode's HTTP API with harness.fetch #2567 / [FEATURE]: sdk: expose the embedded server's fetch handler on PromiseSdk.Interface anomalyco/opencode#54803.
the leases, the agents/streams copy of events and the global fetch patch are not taken. OpenCode already keeps a durable, numbered log (session.log()), and the wake job already covers recovery after eviction.

const stream = await harness.session(id).events();
render(stream.snapshot); // { messages, pending, busy, permissions }
stream.start((events) => {
  for (const event of events) apply(event);
  // run_start, operation_start, message_start, text_delta, tool_start,
  // tool_end, message_end, operation_end { result }, run_end, ...
});
await stream.stop();

events.ts translates OpenCode's typed events. It keeps what they leave out: tool names (session.tool.called has only the call id), and which inputs the current answer is for, so operation_end carries the same result wait() returns.
The stream subscribes before reading the snapshot, and buffers until start(), so nothing falls in between.
Breaking for the experimental API: events() used to be an async iterable of OpenCode's raw events. Those are still available from log().

pnpm exec vitest run --project harness-opencode ran, with new tests for the stream, tool names, queued follow-ups, abort and failure. Ran the example against Workers AI: write and read a file, and clone a repo with exec.

…the example session.events() now returns { snapshot, start, stop } with OpenCodeSessionEvents, the same kinds of event PiHarness and ContainerHarness report, translated from OpenCode's typed events. OpenCode's raw events stay available from session.log(). The example replaces its notes plugin with the Workspace tools from @cloudflare/computer, offered to the model by name. Takes the event union and the Workspace tools from #2483 (Aron).

changeset-bot · 2026-10-12T12:22:26Z

🦋 Changeset detected

Latest commit: 908912e

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 2 packages

Name	Type
agents	Patch
@cloudflare/agent-think	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

devin-ai-integration

Devin Review found 3 potential issues.

devin-ai-integration · 2026-10-12T12:24:44Z

SessionEventProjector

+      const snapshot = await this.#snapshot(opencode);
+      projector = new SessionEventProjector(snapshot);
+      for (const event of early.splice(0)) deliver(projector.project(event));


🔴 Snapshot overlap duplicates operation completions

If delivery occurs during events() snapshot reads, SessionEventProjector can seed and replay the same running operation. The stream then emits two operation_end events for one submission.

Learn more

The stream subscribes before reading session state and buffers incoming OpenCode events in early. The pending read can observe an input already delivered while early still contains its delivery event. SessionEventProjector seeds its running IDs from snapshot.pending and then appends the delivered ID during replay. The same input gets two results at run end. Other overlapping events can also reset the answer reconstructed from the snapshot.

Example: A submit delivers op-1 while events() reads the snapshot. The snapshot includes op-1 as running, and early includes its session.inbox.delivered event. Replaying it appends op-1 twice; the next run end reports op-1 twice.

Recommended fix: Reconcile buffered events with the captured snapshot using a consistent event position, or make replay idempotent against snapshot state, including operation IDs and assistant state. Test delivery and step events interleaved with each snapshot read.

Was this helpful? React with 👍 or 👎 to provide feedback.

Fixed in 908912e. Replay is now idempotent against the snapshot: the projector skips a delivery for an input it already has as running or has already ended, so overlap can't start or end it twice. Unit tests in tests/events.test.ts cover a delivery that repeats the snapshot, and a repeat after the input ended.

agent-think · 2026-10-12T12:26:21Z

🔴 agents import sizes: 1 entry point over threshold

	Entry point	Exports	Largest gzip change	Size now
🔴	`agents/harness/opencode`	2 resized	up to +1.8 KiB (+144.33%)	3 KiB
🟡	`agents/mcp`	15 resized	up to +3.8 KiB (+0.98%)	396.8 KiB
🟡	`agents/mcp/server`	1 resized	+2.3 KiB (+2.8%)	82.8 KiB
🟡	`agents/chat-sdk`	5 resized	up to +1.3 KiB (+0.49%)	267 KiB
🟡	`agents`	30 resized	up to +1.3 KiB (+0.49%)	264.5 KiB
🟡	`agents/workflows`	2 resized	up to +1.3 KiB (+0.49%)	265.8 KiB
🟡	`agents/mcp/client`	1 resized	+1.3 KiB (+0.81%)	160.1 KiB

Changed exports (56)

	Import	Gzip change	Size now
🔴	`agents/harness/opencode#OpenCodeSession`	+1.8 KiB (+144.33%)	3 KiB
🔴	`agents/harness/opencode#OpenCodeHarness`	+1.5 KiB (+26.11%)	7.3 KiB
🟡	`agents/mcp#createMcpHandler`	+3.8 KiB (+0.98%)	396.8 KiB
🟡	`agents/mcp#experimental_createMcpHandler`	+2.3 KiB (+0.61%)	383.1 KiB
🟡	`agents/mcp#createLegacyMcpHandler`	+2.3 KiB (+0.6%)	383 KiB
🟡	`agents/mcp/server#createMcpHandler`	+2.3 KiB (+2.8%)	82.8 KiB
🟡	`agents/mcp#WorkerTransport`	+2.2 KiB (+0.63%)	352.9 KiB
🟡	`agents/mcp#SSEEdgeClientTransport`	+1.9 KiB (+0.54%)	349.2 KiB
🟡	`agents/mcp#StreamableHTTPEdgeClientTransport`	+1.9 KiB (+0.53%)	349.2 KiB
🟡	`agents/mcp#getMcpAuthContext`	+1.8 KiB (+0.53%)	349.1 KiB
🟡	`agents/mcp#MCP_SERVER_ID_MAX_LENGTH`	+1.8 KiB (+0.53%)	349.1 KiB
🟡	`agents/mcp#RPCClientTransport`	+1.8 KiB (+0.53%)	349.1 KiB
🟡	`agents/mcp#normalizeServerId`	+1.8 KiB (+0.53%)	349.1 KiB
🟡	`agents/mcp#RPCServerTransport`	+1.8 KiB (+0.53%)	349.1 KiB
🟡	`agents/mcp#DurableObjectEventStore`	+1.8 KiB (+0.53%)	349.1 KiB
🟡	`agents/mcp#McpAgent`	+1.8 KiB (+0.53%)	349.1 KiB
🟡	`agents/mcp#ElicitRequestSchema`	+1.8 KiB (+0.53%)	349.1 KiB
🟡	`agents/mcp#RPC_DO_PREFIX`	+1.8 KiB (+0.53%)	349.1 KiB
🟡	`agents/chat-sdk#ChatSdkStateAdapter`	+1.3 KiB (+0.49%)	267 KiB
🟡	`agents/chat-sdk#createChatSdkState`	+1.3 KiB (+0.49%)	267.1 KiB
🟡	`agents#getCurrentAgent`	+1.3 KiB (+0.49%)	264.5 KiB
🟡	`agents#__DO_NOT_USE_WILL_BREAK__withInvocationScope`	+1.3 KiB (+0.49%)	264.5 KiB
🟡	`agents#isDurableObjectStorageReset`	+1.3 KiB (+0.49%)	264.5 KiB
🟡	`agents/workflows#AgentWorkflow`	+1.3 KiB (+0.49%)	265.8 KiB
🟡	`agents/chat-sdk#ChatSdkStateAgent`	+1.3 KiB (+0.49%)	266.3 KiB
🟡	`agents/mcp/client#MCPClientManager`	+1.3 KiB (+0.81%)	160.1 KiB
🟡	`agents#unstable_callable`	+1.3 KiB (+0.49%)	264.6 KiB
🟡	`agents#routeAgentRequest`	+1.3 KiB (+0.48%)	265.2 KiB
🟡	`agents#normalizeServerId`	+1.3 KiB (+0.48%)	264.5 KiB
🟡	`agents#isPlatformTransientError`	+1.3 KiB (+0.48%)	264.5 KiB
🟡	`agents#buildAgentUrl`	+1.3 KiB (+0.48%)	265.2 KiB
🟡	`agents#buildAgentPath`	+1.3 KiB (+0.48%)	265 KiB
🟡	`agents#getAgentByName`	+1.3 KiB (+0.48%)	264.4 KiB
🟡	`agents#camelCaseToKebabCase`	+1.3 KiB (+0.48%)	264.4 KiB
🟡	`agents#__DO_NOT_USE_WILL_BREAK__agentContext`	+1.3 KiB (+0.48%)	264.4 KiB
🟡	`agents#DEFAULT_AGENT_STATIC_OPTIONS`	+1.3 KiB (+0.48%)	264.4 KiB
🟡	`agents#isDurableObjectCodeUpdateReset`	+1.3 KiB (+0.48%)	264.4 KiB
🟡	`agents#MCP_SERVER_ID_MAX_LENGTH`	+1.3 KiB (+0.48%)	264.4 KiB
🟡	`agents#SqlError`	+1.3 KiB (+0.48%)	264.4 KiB
🟡	`agents#routeAgentEmail`	+1.3 KiB (+0.48%)	264.7 KiB
🟡	`agents#parseSubAgentPath`	+1.3 KiB (+0.48%)	264.4 KiB
🟡	`agents#isDurableObjectMemoryLimitReset`	+1.3 KiB (+0.48%)	264.4 KiB
🟡	`agents/chat-sdk#defaultThreadShard`	+1.3 KiB (+0.48%)	264.5 KiB
🟡	`agents/chat-sdk#defaultKeyShard`	+1.3 KiB (+0.48%)	264.6 KiB
🟡	`agents#DurableObjectOAuthClientProvider`	+1.3 KiB (+0.48%)	264.4 KiB
🟡	`agents#AGENT_TOOL_MILESTONE_PART`	+1.3 KiB (+0.48%)	264.4 KiB
🟡	`agents#Agent`	+1.3 KiB (+0.47%)	264.4 KiB
🟡	`agents#StreamingResponse`	+1.2 KiB (+0.47%)	264.4 KiB
🟡	`agents#SUB_PREFIX`	+1.2 KiB (+0.47%)	264.4 KiB
🟡	`agents#createHeaderBasedEmailResolver`	+1.2 KiB (+0.47%)	264.6 KiB

…and 6 more exports in the workflow artifact.

How this works

Each runtime export is bundled on its own, minified, and gzipped. Changes smaller than 100 B, or smaller than 1% and 1 KiB, are ignored. Growth over 10% or 5 KiB is marked 🔴. This report is informational and does not fail CI. The workflow artifact contains every measurement.

_{Compared 48d9c36f → 908912ea · workflow run · reported by agent-think[bot]}

… RPC

pkg-pr-new · 2026-10-12T12:54:02Z

Open in StackBlitz

agents

npm i https://pkg.pr.new/agents@2568

@cloudflare/ai-chat

npm i https://pkg.pr.new/@cloudflare/ai-chat@2568

@cloudflare/codemode

npm i https://pkg.pr.new/@cloudflare/codemode@2568

hono-agents

npm i https://pkg.pr.new/hono-agents@2568

@cloudflare/shell

npm i https://pkg.pr.new/@cloudflare/shell@2568

@cloudflare/think

npm i https://pkg.pr.new/@cloudflare/think@2568

@cloudflare/voice

npm i https://pkg.pr.new/@cloudflare/voice@2568

@cloudflare/worker-bundler

npm i https://pkg.pr.new/@cloudflare/worker-bundler@2568

commit: 908912e

… after stop - An input delivered while events() reads the snapshot was both seeded from pending and replayed, so it ended twice. The projector now skips deliveries of inputs it already has or has ended. - The answer is reset when inputs settle and when a new run starts, so a run with no assistant step no longer reports the previous run's text. - start() after stop() delivers nothing, and stop() drops buffered events.

mattzcarey mentioned this pull request Oct 12, 2026

Harness opencode #2483

Closed

devin-ai-integration Bot reviewed Oct 12, 2026

View reviewed changes

test(opencode): return the stream snapshot's transcript as texts over…

7e17991

… RPC

mattzcarey merged commit 46570f5 into main Oct 12, 2026
20 of 21 checks passed

mattzcarey deleted the feat/opencode-session-events branch October 12, 2026 15:51

github-actions Bot mentioned this pull request Oct 12, 2026

Version Packages #2570

Open

Navigation Menu

Repository navigation

feat(agents): OpenCodeHarness session events, and Workspace tools in the example - #2568

feat(agents): OpenCodeHarness session events, and Workspace tools in the example#2568
mattzcarey merged 3 commits into
mainfrom
feat/opencode-session-events

mattzcarey commented Oct 12, 2026 •

edited by devin-ai-integration Bot

Loading

Uh oh!

changeset-bot Bot commented Oct 12, 2026 •

edited

Loading

Uh oh!

devin-ai-integration Bot left a comment

Uh oh!

devin-ai-integration Bot Oct 12, 2026

Uh oh!

mattzcarey Oct 12, 2026

Uh oh!

Uh oh!

Uh oh!

agent-think Bot commented Oct 12, 2026 •

edited

Loading

Uh oh!

pkg-pr-new Bot commented Oct 12, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

mattzcarey commented Oct 12, 2026 • edited by devin-ai-integration Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

changeset-bot Bot commented Oct 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🦋 Changeset detected

Uh oh!

devin-ai-integration Bot left a comment

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Oct 12, 2026

Choose a reason for hiding this comment

Uh oh!

mattzcarey Oct 12, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

agent-think Bot commented Oct 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔴 agents import sizes: 1 entry point over threshold

Uh oh!

pkg-pr-new Bot commented Oct 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

mattzcarey commented Oct 12, 2026 •

edited by devin-ai-integration Bot

Loading

changeset-bot Bot commented Oct 12, 2026 •

edited

Loading

agent-think Bot commented Oct 12, 2026 •

edited

Loading

pkg-pr-new Bot commented Oct 12, 2026 •

edited

Loading