feat: enhance task routing with model-specific strengths and automatic failover

[ianalloway]

feat: enhance task routing with model strengths and auto-failover

Summary

Overhauls the Task Routing Map in AGENTS.md to route tasks to models based on their specific strengths rather than defaulting everything to MiniMax 2.5. Also adds automatic failover rules and a model strengths cheat sheet.

What changed:

• Primary model shift: Claude replaces MiniMax 2.5 as default for reasoning, coding, writing, and job applications
• New task categories: Creative writing, data analysis, long context/docs (→ Gemini), image understanding (→ Gemini), quick Q&A
• Automatic failover rules: Documented behavior for rate limits, auth errors, timeouts, and context overflow — with four ordered failover chains (primary, research, budget, long context)
• Model strengths cheat sheet: Quick-reference table of what each model is best/worst at
• Expanded roadmap: Future multi-agent architecture now includes trading signals agent

What didn't change: Budget ($30/mo), safety rules, memory architecture, autonomy levels, quiet hours, or any other section of AGENTS.md.

Review & Testing Checklist for Human

• Verify Claude is actually configured as an accessible model — the old config was MiniMax 2.5-centric. If Claude isn't set up with sufficient API quota, the bot will immediately hit the failover chain on every request. Check ~/.openclaw/openclaw.json and auth profiles.
• Budget math still works — routing most tasks through Claude ($20/mo sub) plus MiniMax ($10/mo) plus occasional Perplexity/Gemini usage. Confirm this stays under the $30/mo hard limit documented in Spending Limits.
• Failover chains match actual gateway config — these rules are descriptive documentation the bot reads as behavioral guidance. They do NOT configure model.fallbacks in the OpenClaw gateway. Verify the actual gateway fallback config matches these documented chains, or update one to match the other.
• Gemini availability — newly added for long context and vision tasks. Confirm Gemini API key is configured and accessible.

Notes

• This is a documentation/behavioral-config change only — no executable code. The bot reads AGENTS.md at session start to determine how to route tasks.
• There's no automated way to test this. The only verification is observing the bot's behavior in live sessions after merge.

---

Link to Devin run: https://app.devin.ai/sessions/86b10eb8263049ef8e1b94ef22521f81
Requested by: @ianalloway

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring