#26 Support Gemini as a fleet member LLM backend

[Anonymous]

Originally created by: kumaakh

Description

Add support for Google Gemini as an alternative LLM backend for fleet members, alongside Claude.

Motivation

Not all users have access to Claude or may prefer to use Gemini for certain workloads (cost, availability, capability mix). Supporting multiple LLM backends makes the fleet more accessible and flexible. Users should be able to mix and match — e.g., some members running Claude, others running Gemini — within the same fleet.

Scope

MCP Server (src/)

• Member registration should allow specifying the LLM provider/backend
• execute_prompt should route to the appropriate backend CLI/API
• provision_auth should support Gemini API key provisioning (GEMINI_API_KEY)
• update_claude tool needs generalization (update_agent_cli or similar)
• OS command builders (buildPromptCommand, claudeVersion, etc.) need Gemini equivalents
• Error classification (prompt-errors.ts) needs Gemini-specific patterns
• Agent type/interface needs an llmProvider field

PM Skill (skills/pm/)

• Model selection logic (haiku→sonnet→opus escalation) needs Gemini equivalents
• CLAUDE.md templates reference Claude-specific behavior
• Doer-reviewer loop assumes Claude CLI session semantics
• Troubleshooting guide assumes Claude error patterns

Documentation

• All tool docs reference Claude CLI exclusively
• User guide needs multi-provider setup instructions
• Architecture doc needs provider abstraction layer description

Non-goals (v1)

• Gemini as the PM's own LLM (covered by separate provider-for-PM work)
• Automatic provider failover between Gemini and Claude

[Anonymous]

Ticket changed by: kumaakh

• status: open --> closed