summoned-ai-gateway

Add a new cache backend implementation to the Summoned AI Gateway. Trigger whenever the user wants to support a new cache store — e.g. "add Memcached support", "cache to Cloudflare KV", "use DynamoDB as cache", "swap cache for Upstash", "add in-memory fallback". Does NOT trigger for tuning TTL or existing Redis cache (that's a config change).

Add a new guardrail type (input or output validation rule) to the Summoned AI Gateway. Trigger when the user wants a new safety check, content filter, PII detector, prompt-injection detector, toxicity filter, or any per-request validation — e.g. "add a toxicity guardrail", "block prompts with X", "filter output for Y", "add a prompt-injection check".

Add a new Hono middleware to the Summoned AI Gateway. Trigger whenever the user wants to intercept every request — e.g. "add IP allowlist middleware", "add a request logger", "add per-tenant quota middleware", "add a GeoIP-based routing middleware", "add CORS tweaks".

Scaffold a new versioned prompt template for the Summoned AI Gateway and upload it to a running gateway's /admin/prompts API. Trigger whenever the user wants to add/create/store a prompt, save a system prompt, version a template, or "make this prompt reusable" — e.g. "add a customer-support prompt", "save this as a prompt template", "version my prompt for Claude".

Add a new HTTP route or router to the Summoned AI Gateway (Hono). Trigger whenever the user wants a new endpoint, REST resource, admin API, webhook receiver, or WebSocket handler — e.g. "add a /v1/rerank endpoint", "expose an endpoint for X", "we need an API for Y", "add a webhook for Z".

Benchmark one or more providers through the gateway — latency, tokens/second, cost per completion, and streaming first-byte-time. Trigger when the user says "benchmark X", "compare providers", "is provider Y fast enough?", "measure TTFT for …", or wants numbers for a blog/docs post.

Load deep, structured context about the Summoned AI Gateway before making non-trivial changes. Use at the start of a new session when the task touches gateway internals (providers, routers, middlewares, lib/) and CLAUDE.md alone isn't enough. Also trigger on requests like "orient yourself", "give me the lay of the land", "what's in this repo?".

Draft a community RFC (Request for Comments) for a non-trivial change to the Summoned AI Gateway — anything that touches the public contract, core architecture, or cross-cutting concerns. Trigger when the user says "I want to propose X", "let's RFC this", "design doc for Y", or when a change obviously needs broader discussion (breaking header change, new middleware ordering, new top-level config field).

Add a single new feature or field to BOTH first-party SDKs (TypeScript @summoned/ai and Python summoned-ai) in one pass. Trigger when the user says "add X to the SDKs", "expose config option Y in both SDKs", "TS SDK got feature Z, mirror to Python", "add helper for W in the clients". For a full parity audit use the sdk-syncer agent instead.

Update per-token pricing for one or more providers in src/lib/models/*.ts. Trigger when the user says prices changed, a new model is out, cost numbers look wrong, or "refresh pricing for X". For a full autonomous pass across a provider, spawn the pricing-researcher subagent instead.