Auto-detect parallelizable tasks and suggest /fleet mode

[johnpapa]

Feature Request

Summary

Copilot CLI should automatically detect when a user prompt contains parallelizable subtasks and proactively suggest using /fleet mode, rather than requiring the user to know about and manually invoke /fleet.

Problem

Currently, /fleet mode is opt-in via an explicit command. This means:

• Users who don't know about /fleet will always run tasks serially, missing significant speed gains.
• Even users who know about /fleet must manually assess whether their prompt is parallelizable before deciding to use it.
• The CLI already has the intelligence to decompose tasks into subtasks (it does this inside fleet mode) — it just doesn't surface that analysis before execution.

Proposed Solution

When a user submits a prompt (without /fleet), Copilot CLI should:

1. Analyze the prompt for independent, parallelizable subtasks.
2. If parallelizable work is detected, prompt the user:

   "I detected 3 independent tasks here (auth tests, payments tests, notifications tests). Want me to run them in parallel with fleet mode? (y/n)"

3. If the user confirms, execute in fleet mode. If declined, proceed serially as usual.

This keeps the user in control (addressing cost and coordination concerns) while surfacing the optimization opportunity they might otherwise miss.

Why not just always use fleet?

Fleet mode has real tradeoffs that justify keeping it opt-in:

• Cost: Each subagent consumes its own token budget — fleet can multiply usage significantly.
• Coordination risk: Parallel agents editing files concurrently can cause merge conflicts, duplicate code, or inconsistent patterns.
• Not everything is parallelizable: Many prompts are inherently sequential. Auto-detecting true independence vs. hidden dependencies is non-trivial — getting it wrong produces broken code.
• Determinism: Serial execution is more predictable and easier to debug.

The "suggest, don't assume" approach addresses all of these — the user gets the benefit of the CLI's analysis without losing control.

Expected UX Flow

> Add unit tests for the auth module, payments module, and notifications module

🔀 I detected 3 independent subtasks that could run in parallel with /fleet.
   This would be ~3x faster but uses more tokens. Run in fleet mode? [y/N]

Alternatives Considered

• Always auto-fleet: Too aggressive — cost and correctness concerns.
• Status quo (manual /fleet only): Leaves optimization opportunity hidden from most users.
• Config flag to always fleet: Possible, but per-prompt suggestion is more flexible.

Additional Context

This feels like a natural evolution of fleet mode — ship the explicit command first (done ✅), then add smart suggestions on top.

---

Moved from cli/cli#12837 (closed as not_planned — wrong repo).