Cookbook recipe + skills directory + Broadcast Destinations: AgentGuard Spend integration

[MerchantGuardOps]

Issue title

Add cookbook recipe: Per-task spend caps + signed audit receipts with AgentGuard Spend

---

Issue body (paste into github.com/OpenRouterTeam/docs/issues/new)

Hi team,

I'm JP Montgomery, founder of AgentGuard. We just shipped a zero-data-plane SDK (@agentguard-run/spend on npm, agentguard-spend on PyPI) that wraps OpenAI-compatible clients with per-task spend caps, capability gates, and Ed25519-signed audit receipts.

OpenRouter is our headline integration path — our docs at https://agentguard.run/docs/openrouter-quickstart lead with the pattern (baseURL: 'https://openrouter.ai/api/v1') because your unified-API + one-key model is the cleanest way for a customer's CFO to assign which models which teams can use for which tasks.

I'd love to contribute a cookbook recipe covering this pattern. Three angles I think would fit alongside your existing recipes:

1. Per-task spend caps with signed audit receipts — uses withSpendGuard() on an OpenRouter-pointed OpenAI client; shows downgrade-on-cap-hit (opus → haiku at $50/day), block-on-ceiling ($200/day), and a verifiable Ed25519 chain for procurement review.

2. Multi-team model routing with CFO-assigned caps — Stripe-Projects-adjacent: shows how to express "engineering team can use claude-opus, support team gets gpt-5-mini, payment-execute tasks require capability claim" as a single SpendPolicy against OpenRouter.

3. Cryptographic evidence preservation for AI agent decisions — fits next to your red-teaming and distillation-compliance recipes; shows the receipt format and a CLI verifier (agentguard verify --trace) plus a browser verifier at https://agentguard.run/verify.

Happy to draft any or all of these as a PR if you'll accept them. Alternatively, if the docs repo is read-only and you only want to point at external content, I'll just host the recipe at https://agentguard.run/docs/cookbook/openrouter-spend-caps and you could link to it.

I notice the Hermes Agent recipe (May 11) is the precedent for external contributors. Let me know what you'd prefer.

Three quick questions tied to this:

• Skills directory: I see OpenRouterTeam/skills and agent-skills repos. We have a SKILL.md ready (name: agentguard-spend). Is there a third-party submission path, or is your skills repo first-party-only? If first-party-only, no problem — we'll publish at MerchantGuardOps/agentguard-skills with our own plugin marketplace.
• awesome-openrouter: preparing a PR adding apps/agentguard-spend/ (app.yaml + logo.png). Anything specific I should know about the review queue?
• Broadcast Destinations: you list Langfuse, Sentry, PostHog, S3, OTel as observability sinks. Would AgentGuard fit as a "tamper-evident audit sink" alongside those? Happy to provide the receipt-emission spec.

Thanks for building OpenRouter. It's the cleanest model-routing layer we've integrated against.

— JP Montgomery
agentguard.run · invest@agentguard.run