[Feature]PlanOnly mode

[CodingBangboo]

What

Adds a planOnly option to runTeam() that lets the coordinator decompose a goal into tasks without executing any task agents. The result has success: true, planOnly: true, and tasks populated with pending/blocked task records: useful for integration testing the task DAG or previewing cost before a real run.

What changed

• src/types.ts — new RunTeamOptions interface (replaces the inline anonymous options type on runTeam()); adds planOnly?: boolean to both RunTeamOptions and TeamRunResult
• src/orchestrator/orchestrator.ts — runTeam() now accepts RunTeamOptions; bypasses the simple-goal short circuit when planOnly: true; inserts a short-circuit after onPlanReady fires (but before executeQueue) that returns success: true with tasks from queue.list()
• src/index.ts — exports RunTeamOptions as part of the public API
• tests/orchestrator.test.ts — four new test cases covering: plan returned without executing tasks, onPlanReady still fires, onPlanReady rejection wins over planOnly, and simple-goal short-circuit is bypassed

Behavior notes

• onPlanReady still fires when planOnly: true. If the callback returns false, the rejection wins — result is success: false and planOnly is undefined (same as today's abort behavior).
• planOnly is a per-call option on the third argument, not on OrchestratorConfig: consistent with the scope guidance in the issue.
• The simple-goal short-circuit is bypassed so the coordinator always runs, which means planOnly on a single-agent team is valid and returns a single-task plan.

Why

fixed #192

[JackChen-me]

@codex review plz

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Delightful!

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".

[JackChen-me]

Approach looks right, but there is one real lifecycle bug.

P1 (must fix): agent_start fires for the coordinator before decomposition, but the new planOnly path returns before the matching agent_complete, which currently only fires after synthesis. Anyone tracking active agents via onProgress will see the coordinator as permanently running for successful plan-only runs. Small fix: emit agent_complete for the decomposition result before the planOnly return, and add a regression test that agent_start / agent_complete stay balanced.

P2 (optional, can be follow-up): renderLiveOutput in the dashboard treats only completed, failed, skipped, and blocked as terminal. In a plan-only result, root tasks intentionally remain pending, so the dashboard header says "Task graph execution in progress" instead of showing this as a completed plan preview. This is outside the main runTeam path, so I would not block on it, but it is worth fixing for callers who render plan-only results.

[JackChen-me]

P1 fixed. P2 is non-blocking / follow-up. Good to merge.