feat: add Agent Team support

[shadowpigy]

Summary

• Enable experimental agent teams for all ACP sessions (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1)
• Forward task_started/task_progress/task_notification as agent_message_chunk notifications
• After result with team activity, continue reading generator with timeout for follow-up messages (30s initial, 5s for subsequent cleanup turns)
• Display TeamCreate/TeamDelete/SendMessage in tool info
• Carry over pending generator promise across prompt() calls via session.pendingNext

Partially addresses #336

Test plan

• npm run lint passes
• npm run format:check passes
• npm run test:run — 126 tests pass
• Manual E2E: direct stdio test with team creation, greeter agent, message exchange, cleanup

🤖 Generated with Claude Code

[cla-bot]

Thank you for your pull request and welcome to our community. We could not parse the GitHub identity of the following contributors: hexarrior.
This is most likely caused by a git client misconfiguration; please make sure to:

1. check if your git client is configured with an email to sign commits git config --list | grep email
2. If not, set it up using git config --global user.email email@example.com
3. Make sure that the git commit email is configured in your GitHub account settings, see https://github.com/settings/emails

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4856710b85

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[cla-bot]

We require contributors to sign our Contributor License Agreement, and we don't have @shadowpigy on file. You can sign our CLA at https://zed.dev/cla. Once you've signed, post a comment here that says '@cla-bot check'.

[cla-bot]

We require contributors to sign our Contributor License Agreement, and we don't have @shadowpigy on file. You can sign our CLA at https://zed.dev/cla. Once you've signed, post a comment here that says '@cla-bot check'.

[shadowpigy]

@cla-bot check

[shadowpigy]

@cla-bot check

[cla-bot]

The cla-bot has been summoned, and re-checked this pull request!

[shadowpigy]

Addressed all review feedback:

• Removed the hardcoded CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS env var — callers now opt-in via _meta.claudeCode.options.env (per @josevalim's suggestion)
• Added cancellation check during team idle timeout (Codex P2)
• Preserved promptResult in replay-handoff path (Codex P1)
• Fixed replaySessionHistory to filter CLI-injected synthetic user messages (teammate-message, system-reminder, etc.) — applied the same filtering logic already used in the live streaming path

[shadowpigy]

Bug found: replaySessionHistory() filter is too broad — drops all text-only user messages

The filter added in replaySessionHistory() to skip CLI-injected synthetic messages (teammate-message, system-reminder, etc.) inadvertently skips all text-only user messages:

if (
  role === "user" &&
  (typeof content === "string" ||
    (Array.isArray(content) && content.length === 1 && content[0].type === "text"))
) {
  continue;
}

The problem is that synthetic and real user text messages are structurally identical — there's no metadata to distinguish them. This causes session/load to replay history without any of the user's text messages (only image messages survive the filter).

Possible fixes:

1. Check content for synthetic markers (e.g. <teammate-message>, <system-reminder> tags)
2. Have CLI add metadata (e.g. _synthetic: true) when injecting synthetic messages
3. Move filtering to the CLI layer (handleUserMessageChunk) instead of ACP

@josevalim What's your take on the right approach here?