pick-agent-stacklisted

# Pick Agent Stack ## Overview `pick-gtm-agent-pattern` decides an agent's **shape** (pattern, tier, flowchart). This skill decides **what runs it**: the execution engine, LLM routing, evals, observability, state, secrets, and ingress — chosen per layer from the mapped core stack, sized to the agent's tier so a simple report agent doesn't get a ten-layer platform. **The recall dataset lives in ONE place — fetch it, never reproduce from memory:** ``` https://labs.bowtiefunnel.com/tools/tools.json ``` Filter to `category === "Agent Infrastructure"` (`core_stack: true`) — the Headless GTM OS components. (Inside the `bowtie-funnel-Labs` repo, read `docs/tools/tools.json` directly.) ## The layer map Layers in pipeline order. An agent uses a layer only if it needs it: | # | Layer | Component | |---|-------|-----------| | 0 | Secrets & config | Infisical | | 1 | Code, review & CI quality gate | GitHub (+ DeepEval PR evals) | | 2 | Durable execution | Trigger.dev | | 3 | Webhook ingress & ACK buffer | Svix | | 4 | Schema validation | Zod / Drizzle | | 5 | PII / secret redaction | CloakPipe | | 6 | LLM routing & fallback | Vercel AI Gateway → OpenRouter (tier-route locally with `tools/llm-switchboard`) | | 7 | LLM observability & cost | Langfuse | | 8 | State, storage & locks | Supabase | | 9 | Human-in-the-loop interface | Slack | ## Steps 1. **Start from the pattern verdict.** If `pick-gtm-agent-pattern` hasn't run, run it first — the tier drives how much stack the agen