菜单
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.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
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.
| name | update-pricing |
| description | 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. |
Pricing lives in src/lib/models/<provider>.ts as ModelDefinition[]. The
gateway reports $0 for unknown models — so out-of-date files quietly
under-report cost.
groq → src/lib/models/groq.ts)Read src/lib/models/<id>.ts
Note existing entries. Never blow away the file.
Use Edit with a narrow old_string/new_string per model. Preserve:
contextWindow, capabilities, descriptiondeprecated: true flagsConvert units correctly: a page quoting "$X per 1K" → X * 1000 in
inputPricePer1M. Cached input rates go in a code comment, NOT in the
primary price field.
Append to the array with the same field order as existing entries. Every
new model needs id, name, inputPricePer1M, outputPricePer1M. Prefer
adding contextWindow and capabilities if documented.
Do NOT delete — mark deprecated: true. Historical request_logs may still
reference the id.
bun run check-types
bun test tests/pricing.test.ts
Then spot-check by computing cost for a known prompt:
curl -s http://localhost:4000/v1/chat/completions \
-H "Authorization: Bearer sk-smnd-..." \
-d '{"model":"<provider>/<model>","messages":[{"role":"user","content":"hi"}]}' \
-D - | grep X-Summoned-Cost-USD
src/lib/models/index.ts — but that path really belongs to the
add-provider skill.