#46 PM skill: increase timeouts for Gemini members, block Gemini as orchestrator

[Anonymous]

Originally created by: kumaakh

Problem

Gemini Flash is considerably slower than Claude for code tasks. The PM skill currently uses uniform timeouts regardless of the member's LLM backend, causing Gemini members to time out on tasks that Claude members complete easily.

Additionally, Gemini has no concept of background agents — every remote prompt blocks. This makes Gemini a poor fit as an orchestrator since the PM needs to dispatch to multiple members concurrently (prep reviewer while doer works, monitor tasks in parallel, etc.).

Requirements

1. Timeout scaling for Gemini members

• PM skill should detect the member's LLM provider and apply a timeout multiplier
• Suggested: 2–3x timeout for Gemini Flash members vs Claude members
• Apply to execute_prompt and execute_command dispatches

2. Warn (not block) if Gemini is used as orchestrator

• PM skill should emit a warning if the orchestrator itself is running on Gemini
• Every execute_prompt call blocks for Gemini — no parallel dispatch possible
• This is the strongest reason Gemini is a poor fit for the PM role
• User's choice is final — if they proceed despite the warning, respect it

3. Allow multiple Gemini members as doers

• Multiple Flash members and multiple Sonnet/Haiku members should be supported as doers
• The limitation is orchestrator-only — Gemini is fine as a worker bee

Context

This came from real-world observation: Gemini Flash is noticeably slower and its blocking nature serializes all fleet operations when used as orchestrator.

[Anonymous]

Originally posted by: kumaakh

Addressed in skills/pm/SKILL.md (main). The Provider Awareness table documents '2-3x timeout multiplier for execute_prompt dispatches to Gemini members' and notes that Gemini does not support background agents. Closing as skill-documented.

[Anonymous]

Ticket changed by: kumaakh

• status: open --> closed