[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6DEJDPP9MQ2NE49VP942KFK":3,"skill-same:01M6DEJDPP9MQ2NE49VP942KFK":114},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":37,"linkOnly":38,"markdownContent":39,"markdownBody":40,"frontMatter":41,"refName":42,"commitSha":43,"skillDirectory":44,"forksCount":14,"repositoryHtmlUrl":45,"repositoryHomepageUrl":6,"ownerHtmlUrl":46,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":47,"install":48,"resources":61},"01M6DEJDPP9MQ2NE49VP942KFK","openrouter-routing",null,"Route cheap grunt work through OpenRouter as a spend-capped CLI\u002FCodex sidecar. Use for T0\u002FT1 model choice, :floor\u002F:nitro\u002F:exacto, spend caps, and generation ids. NEVER Cursor BYOK or Claude ANTHROPIC_BASE_URL.","Prekzursil\u002Fagent-skills-toolchain","Prekzursil","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F54636077?v=4","01M6DACZ1PCQHD99SCR1YXG28B","Andrei Mihai Visalon","01M6DACZ670WZF535V3WYYSS4Z",0,"MIT","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fopenrouter-routing\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36],"openrouter routing","routing","openrouter","cursor","as","generation","route","base","never","use","byok","cli","work","nitro","choice","model","ids","codex","2026-11-10 10:12:30",false,"---\nname: openrouter-routing\ndescription: Route cheap grunt work through OpenRouter as a spend-capped CLI\u002FCodex sidecar. Use for T0\u002FT1 model choice, :floor\u002F:nitro\u002F:exacto, spend caps, and generation ids. NEVER Cursor BYOK or Claude ANTHROPIC_BASE_URL.\n---\n\n# OpenRouter routing (sidecar only)\n\nOpenRouter is a hosted OpenAI-compatible API at `https:\u002F\u002Fopenrouter.ai\u002Fapi\u002Fv1`. It is overflow for CLI\u002FCodex, not a local gateway and not a second Cursor stack. Do not install LiteLLM. Do not add an OpenRouter MCP.\n\n## Key location (name only)\n\n| Where | Role |\n| --- | --- |\n| Environment variable `OPENROUTER_API_KEY` (User and Machine) | Canonical secret |\n| `.agents\u002Flocal.json` field `env_keys.OPENROUTER_API_KEY.env_key` | Pointer by name only |\n| Cursor Models \u002F OpenAI API Key \u002F Override Base URL | Forbidden |\n| `agents.json` MCP registry \u002F `.cursor\u002Fmcp.json` | Do not add OpenRouter |\n\nNever print, commit, or interpolate the secret into logs, skills, or sync output. The T0 script reads the environment variable at runtime.\n\n## Never Cursor BYOK\n\nDo not put the OpenRouter secret into Cursor. The vendor cookbook sets Cursor's OpenAI key plus Override Base URL to `https:\u002F\u002Fopenrouter.ai\u002Fapi\u002Fv1\u002Fcursor`. That can break native Max models; Auto \u002F Composer 2 may ignore BYOK. Composer 2.5 is Cursor-only (not on OpenRouter). Keep Cursor Max exclusive inside Cursor.\n\n## T0 \u002F T1 \u002F T2\n\nEscalate one-way. A cheap outage must not silently spend frontier. No mid-tool-call fallback up the tier (retrying a failed cheap call on a dearer model inside the same tool step costs more than Max).\n\n| Tier | Use | Where |\n| --- | --- | --- |\n| **T0 grunt** | Extract, reformat, bulk classify | Cheapest capable OpenRouter. Never Cursor Max. |\n| **T1 standard** | Drafts, single-file, summaries | Mid OpenRouter. At most one different-vendor fallback for outages, not mid-tool-call retries up-tier. |\n| **T2 frontier** | Architecture, multi-file, adversarial, gates | Cursor Max \u002F Claude Max unchanged. |\n\n- T0 to T1 on schema fail or two retries (new step, allowlisted model).\n- T1 to T2 on correctness-critical scope.\n- Default T0 slug (**paid**, when account balance is above zero): `deepseek\u002Fdeepseek-v4-flash-0731` (verify live). Prefer `:floor` when shopping on price.\n- Cheaper live alias `deepseek\u002Fdeepseek-v4-flash-latest` ($0.03\u002F$0.075 per M) vs the pin `deepseek\u002Fdeepseek-v4-flash-0731` ($0.06\u002F$0.12). The pin is deliberate (stability); the alias is cheaper but moving — verify the slug live before adopting.\n- Default T0 **test** slug (sidecar smoke while paid returns 402): `liquid\u002Flfm-2.5-2.6b:free`. See [Free \u002F test slugs](#free--test-slugs).\n\nAllowlist the model before the call. Do not use OpenRouter's `models` array as a hidden upgrade path. Do not pick a random `:free` slug for agentic coding.\n\n## Suffixes\n\nAppend to a model slug ([provider routing](https:\u002F\u002Fopenrouter.ai\u002Fdocs\u002Fguides\u002Frouting\u002Fprovider-selection), [exacto](https:\u002F\u002Fopenrouter.ai\u002Fdocs\u002Fguides\u002Frouting\u002Fmodel-variants\u002Fexacto)):\n\n| Suffix | Meaning |\n| --- | --- |\n| `:floor` | Cheapest provider (`provider.sort` = price) **and admits flex service tiers** |\n| `:nitro` | Highest throughput (`provider.sort` = throughput) and admits priority\u002Ffast tiers |\n| `:exacto` | Quality-first provider order for tool-calling reliability |\n| `:online` | **Deprecated.** Do not use. Web search is the OpenRouter web-search server tool. |\n\nExplicit `provider.sort` overrides a suffix when both are set. `:floor`\u002F`:nitro` are supersets of `provider.sort` (sort plus service-tier admission), not aliases.\n\n## Batch (bulk T0)\n\n`POST \u002Fapi\u002Fbeta\u002Fbatches` runs latency-insensitive bulk grunt (classification, extraction) at roughly 50% of token price — text-only, ~24h completion window. It is **not** in OpenAPI v1 and still requires a positive credit balance. Use ad hoc for bulk T0 jobs only; do **not** wire a daemon around it.\n\n## Claude Code footgun\n\nDo not set `ANTHROPIC_BASE_URL` to `https:\u002F\u002Fopenrouter.ai\u002Fapi` (that host is not `\u002Fv1`) and do not empty `ANTHROPIC_API_KEY` to point Claude at OpenRouter. That rebills Claude usage to OpenRouter and bypasses Claude Max. Leave Claude on native Max.\n\n## Codex: named provider only\n\nWave 1 merges a **non-default** `[model_providers.openrouter]` into `~\u002F.codex\u002Fconfig.toml`:\n\n- `base_url = \"https:\u002F\u002Fopenrouter.ai\u002Fapi\u002Fv1\"`\n- `env_key = \"OPENROUTER_API_KEY\"`\n- `wire_api = \"responses\"` (Codex no longer accepts `chat`)\n\nDefault Codex `model` stays ChatGPT-native. Do not make OpenRouter the default.\n\nOpt-in: `codex --profile openrouter`, or set `model_provider = \"openrouter\"` for that invocation only.\n\nOpenRouter serves `\u002Fapi\u002Fv1\u002Fresponses` (Codex wire). The T0 grunt script uses `\u002Fapi\u002Fv1\u002Fchat\u002Fcompletions`.\n\n## Spend-cap checklist (hard daily + monthly)\n\nAgents cannot operate the OpenRouter dashboard. A normal inference credential can **read** limits via `GET \u002Fapi\u002Fv1\u002Fkey` (`limit`, `limit_reset`, `limit_remaining`, `usage_daily`, `usage_monthly`). Changing limits needs the dashboard or a Management API credential (`PATCH \u002Fapi\u002Fv1\u002Fkeys\u002F{hash}`). Management credentials cannot run completions. Do not mint new secrets from an agent session.\n\nOne inference credential has **one** `limit_reset`: `daily`, `weekly`, `monthly`, or none — not daily and monthly on the same credential.\n\n### A. Per-key cap (free \u002F PAYG) — you click\n\n1. Open [https:\u002F\u002Fopenrouter.ai\u002Fsettings\u002Fkeys](https:\u002F\u002Fopenrouter.ai\u002Fsettings\u002Fkeys) (same list: [https:\u002F\u002Fopenrouter.ai\u002Fkeys](https:\u002F\u002Fopenrouter.ai\u002Fkeys)).\n2. Click the **existing** inference key. Do not recreate it. Do not paste it into Cursor.\n3. Set **Limit** to a USD credit cap for this key.\n4. Set **Limit reset** to **Daily** (midnight UTC) for a hard daily brake, **or** **Monthly** for a monthly brake — pick one.\n5. Save. Confirm with `GET \u002Fapi\u002Fv1\u002Fkey` (script `--show-limits`).\n\nTo approximate daily + monthly on PAYG without Enterprise: set the key to **daily** (hard brake) and watch `usage_monthly` from `GET \u002Fapi\u002Fv1\u002Fkey`. Monthly is then observational unless you add B.\n\n### B. Daily AND monthly hard caps together — you click\n\nWorkspace budgets can set both intervals (strictly decreasing: monthly greater than daily). Dashboard:\n\n1. Open `https:\u002F\u002Fopenrouter.ai\u002Fworkspaces\u002F\u003Cslug>\u002Fsettings` (Default workspace is allowed).\n2. Scroll to **Budgets**.\n3. Click **Add budget**, choose interval `daily`, enter a dollar amount; repeat for `monthly` (monthly must be greater than daily).\n4. Click **Save**. Organization Administrator role required. Workspace budgets are documented as Enterprise.\n\nAPI equivalent (management credential only): `PUT \u002Fapi\u002Fv1\u002Fworkspaces\u002F{id}\u002Fbudgets\u002Fdaily` and `...\u002Fmonthly` with a `limit_usd` body. Create a management credential only in the browser at [https:\u002F\u002Fopenrouter.ai\u002Fsettings\u002Fmanagement-keys](https:\u002F\u002Fopenrouter.ai\u002Fsettings\u002Fmanagement-keys).\n\nIf `GET \u002Fapi\u002Fv1\u002Fkey` shows `limit` of null, the per-key cap is not set. Treat dashboard cap as **NOT-CHECKED** until you complete A (and B if you need both windows).\n\n## Free \u002F test slugs\n\nPaid T0 (`deepseek\u002Fdeepseek-v4-flash-0731`) returned HTTP 402 on this account while slightly overdrawn. Free variants are for **sidecar smoke and cheap extract**, not agentic coding. Allowlist before the call. Catalog snapshot: [openrouter-free-models_2026-08-27.md](openrouter-free-models_2026-08-27.md).\n\nLive `GET \u002Fapi\u002Fv1\u002Fmodels` on 2026-08-27: 417 models, **20** with `pricing.prompt` and `pricing.completion` of `0` (17 `:free` chat slugs, router `openrouter\u002Ffree`, plus two Lyria audio previews that are not chat T0). Re-fetch before treating this list as current.\n\n| Role | Slug | Live result 2026-08-27 |\n| --- | --- | --- |\n| **T0 test** (smoke \u002F pong \u002F extract) | `liquid\u002Flfm-2.5-2.6b:free` | HTTP **200**. `GET \u002Fgeneration` later `total_cost=0` (provider Liquid). Liquid advises against agentic coding and may train on free traffic. |\n| **T0 test router** (random free; not for coding) | `openrouter\u002Ffree` | HTTP **200**, resolved to `nvidia\u002Fnemotron-3.5-lightning:free`. `GET \u002Fgeneration` later `total_cost=0` (Nvidia). |\n| Not a credit failure | `google\u002Fgemma-4-26b-a4b-it:free` | HTTP **429** upstream rate-limit (`Provider returned error`), not 402. |\n| **Paid cheap** (when credits exist) | `deepseek\u002Fdeepseek-v4-flash-0731` plus `:floor` | HTTP **402** Insufficient credits. Paid 3-model bake-off stays skipped until the balance is above zero. |\n\nOfficial docs say a **negative account balance may 402 even free models**. This account's live test: `:free` returned 200 while paid DeepSeek 402'd. Do not claim that is true for every overdraft; if `:free` starts 402, add credits — that is not a skill lie.\n\nPlatform caps on IDs ending in `:free`: 20 requests\u002Fminute; 50 requests\u002Fday if lifetime purchases are under $10, else 1000\u002Fday. This inference key has `is_free_tier=false` and per-key `limit=null` (dashboard Limit still a user click). `GET \u002Fapi\u002Fv1\u002Fcredits` needs `OPENROUTER_MANAGEMENT_KEY` (absent).\n\nFree vs paid training toggles are separate in [privacy settings](https:\u002F\u002Fopenrouter.ai\u002Fsettings\u002Fprivacy). Poolside `:free` pages state they may train on free inputs\u002Foutputs. Not for secrets. Do not use `:online` (deprecated) or OpenRouter web search in place of Firecrawl\u002FExa\u002Fnative Max.\n\n## Low-refusal research lane (allowlisted)\n\n`nousresearch\u002Fhermes-4-70b` — $0.13\u002F$0.40 per M, 131K ctx, reasoning-capable, RefusalBench SOTA. Lane: low-refusal\u002Fsteerable research **chat\u002Fextraction\u002Fsynthesis only** — e.g. pharmacology\u002Fharm-reduction literature queries the default hosted lanes over-refuse. Allowlist before the call, as with every slug.\n\nHARD constraints:\n\n- **No tool calling** — the OpenRouter endpoint does not expose `tools`. Never route agentic T0\u002FT1 (grunt-with-tools or T1 tool work) here.\n- **Single provider (Nebius)** — no fallback diversity; provider-sort suffixes are no-ops.\n- **Bare slug only** — no `:free` variant exists; do not append `:floor`\u002F`:nitro`\u002F`:exacto`.\n- **Paid** — 402s while the account is overdrawn; waits on the credit top-up like the other paid slugs.\n- `nousresearch\u002Fhermes-4-405b` ($1.00\u002F$3.00) is **escalation-only** for correctness-critical steerable work, per the T0→T1 escalation rules.\n\nEvidence: [openrouter-apps-hermes-uncensored_2026-08-27.md](openrouter-apps-hermes-uncensored_2026-08-27.md).\n\n## Probe commands\n\nThin T0 script: reads `OPENROUTER_API_KEY` from the environment, sends one chat completion with app headers `HTTP-Referer` and `X-OpenRouter-Title` (`agents-openrouter-sidecar`), prints generation id plus usage, then reads spend from `GET \u002Fapi\u002Fv1\u002Fgeneration?id=...` when an id is present. It never prints the secret. `GET \u002Fgeneration` can 404 for a few seconds; retry.\n\n```powershell\n# Wave 0 smoke while paid T0 is 402:\nC:\\Python314\\python.exe C:\\Users\\Prekzursil\\.agents\\scripts\\openrouter-complete.py --model liquid\u002Flfm-2.5-2.6b:free --prompt \"pong\"\n\nC:\\Python314\\python.exe C:\\Users\\Prekzursil\\.agents\\scripts\\openrouter-complete.py --show-limits --model liquid\u002Flfm-2.5-2.6b:free --prompt \"pong\"\n\n# Paid cheap (only after credits exist):\nC:\\Python314\\python.exe C:\\Users\\Prekzursil\\.agents\\scripts\\openrouter-complete.py --prompt \"Reply with the single word: pong\"\n```\n\nDo not run verbose HTTP clients against these endpoints; they can echo auth headers.\n\n## Noted, not wired (ADOPT-later)\n\n- `openrouter:web_search` server tool exists (beta) — the `:online` replacement **if ever** needed; search stays on Firecrawl\u002FMCP per routing decision.\n- Embeddings + rerank live on a **separate** catalog (`GET \u002Fapi\u002Fv1\u002Fembeddings\u002Fmodels`, 33 models, not in `\u002Fmodels`) — ADOPT-later, nothing wired.\n- `session_id` sticky routing for prompt-cache hits (~10 min stickiness) — ADOPT-later.\n\n## Related\n\n- `cost-tracking` — OpenRouter generation ids plus the Claude SQLite tracker (label the source).\n- `cost-aware-llm-pipeline` — generic routing patterns (not OpenRouter SSOT).\n- Evidence: [openrouter-complete-surface_2026-08-27.md](openrouter-complete-surface_2026-08-27.md) (38-source full-surface audit) and [openrouter-free-models_2026-08-27.md](openrouter-free-models_2026-08-27.md) (free-slug catalog snapshot).\n\n","# OpenRouter routing (sidecar only)\n\nOpenRouter is a hosted OpenAI-compatible API at `https:\u002F\u002Fopenrouter.ai\u002Fapi\u002Fv1`. It is overflow for CLI\u002FCodex, not a local gateway and not a second Cursor stack. Do not install LiteLLM. Do not add an OpenRouter MCP.\n\n## Key location (name only)\n\n| Where | Role |\n| --- | --- |\n| Environment variable `OPENROUTER_API_KEY` (User and Machine) | Canonical secret |\n| `.agents\u002Flocal.json` field `env_keys.OPENROUTER_API_KEY.env_key` | Pointer by name only |\n| Cursor Models \u002F OpenAI API Key \u002F Override Base URL | Forbidden |\n| `agents.json` MCP registry \u002F `.cursor\u002Fmcp.json` | Do not add OpenRouter |\n\nNever print, commit, or interpolate the secret into logs, skills, or sync output. The T0 script reads the environment variable at runtime.\n\n## Never Cursor BYOK\n\nDo not put the OpenRouter secret into Cursor. The vendor cookbook sets Cursor's OpenAI key plus Override Base URL to `https:\u002F\u002Fopenrouter.ai\u002Fapi\u002Fv1\u002Fcursor`. That can break native Max models; Auto \u002F Composer 2 may ignore BYOK. Composer 2.5 is Cursor-only (not on OpenRouter). Keep Cursor Max exclusive inside Cursor.\n\n## T0 \u002F T1 \u002F T2\n\nEscalate one-way. A cheap outage must not silently spend frontier. No mid-tool-call fallback up the tier (retrying a failed cheap call on a dearer model inside the same tool step costs more than Max).\n\n| Tier | Use | Where |\n| --- | --- | --- |\n| **T0 grunt** | Extract, reformat, bulk classify | Cheapest capable OpenRouter. Never Cursor Max. |\n| **T1 standard** | Drafts, single-file, summaries | Mid OpenRouter. At most one different-vendor fallback for outages, not mid-tool-call retries up-tier. |\n| **T2 frontier** | Architecture, multi-file, adversarial, gates | Cursor Max \u002F Claude Max unchanged. |\n\n- T0 to T1 on schema fail or two retries (new step, allowlisted model).\n- T1 to T2 on correctness-critical scope.\n- Default T0 slug (**paid**, when account balance is above zero): `deepseek\u002Fdeepseek-v4-flash-0731` (verify live). Prefer `:floor` when shopping on price.\n- Cheaper live alias `deepseek\u002Fdeepseek-v4-flash-latest` ($0.03\u002F$0.075 per M) vs the pin `deepseek\u002Fdeepseek-v4-flash-0731` ($0.06\u002F$0.12). The pin is deliberate (stability); the alias is cheaper but moving — verify the slug live before adopting.\n- Default T0 **test** slug (sidecar smoke while paid returns 402): `liquid\u002Flfm-2.5-2.6b:free`. See [Free \u002F test slugs](#free--test-slugs).\n\nAllowlist the model before the call. Do not use OpenRouter's `models` array as a hidden upgrade path. Do not pick a random `:free` slug for agentic coding.\n\n## Suffixes\n\nAppend to a model slug ([provider routing](https:\u002F\u002Fopenrouter.ai\u002Fdocs\u002Fguides\u002Frouting\u002Fprovider-selection), [exacto](https:\u002F\u002Fopenrouter.ai\u002Fdocs\u002Fguides\u002Frouting\u002Fmodel-variants\u002Fexacto)):\n\n| Suffix | Meaning |\n| --- | --- |\n| `:floor` | Cheapest provider (`provider.sort` = price) **and admits flex service tiers** |\n| `:nitro` | Highest throughput (`provider.sort` = throughput) and admits priority\u002Ffast tiers |\n| `:exacto` | Quality-first provider order for tool-calling reliability |\n| `:online` | **Deprecated.** Do not use. Web search is the OpenRouter web-search server tool. |\n\nExplicit `provider.sort` overrides a suffix when both are set. `:floor`\u002F`:nitro` are supersets of `provider.sort` (sort plus service-tier admission), not aliases.\n\n## Batch (bulk T0)\n\n`POST \u002Fapi\u002Fbeta\u002Fbatches` runs latency-insensitive bulk grunt (classification, extraction) at roughly 50% of token price — text-only, ~24h completion window. It is **not** in OpenAPI v1 and still requires a positive credit balance. Use ad hoc for bulk T0 jobs only; do **not** wire a daemon around it.\n\n## Claude Code footgun\n\nDo not set `ANTHROPIC_BASE_URL` to `https:\u002F\u002Fopenrouter.ai\u002Fapi` (that host is not `\u002Fv1`) and do not empty `ANTHROPIC_API_KEY` to point Claude at OpenRouter. That rebills Claude usage to OpenRouter and bypasses Claude Max. Leave Claude on native Max.\n\n## Codex: named provider only\n\nWave 1 merges a **non-default** `[model_providers.openrouter]` into `~\u002F.codex\u002Fconfig.toml`:\n\n- `base_url = \"https:\u002F\u002Fopenrouter.ai\u002Fapi\u002Fv1\"`\n- `env_key = \"OPENROUTER_API_KEY\"`\n- `wire_api = \"responses\"` (Codex no longer accepts `chat`)\n\nDefault Codex `model` stays ChatGPT-native. Do not make OpenRouter the default.\n\nOpt-in: `codex --profile openrouter`, or set `model_provider = \"openrouter\"` for that invocation only.\n\nOpenRouter serves `\u002Fapi\u002Fv1\u002Fresponses` (Codex wire). The T0 grunt script uses `\u002Fapi\u002Fv1\u002Fchat\u002Fcompletions`.\n\n## Spend-cap checklist (hard daily + monthly)\n\nAgents cannot operate the OpenRouter dashboard. A normal inference credential can **read** limits via `GET \u002Fapi\u002Fv1\u002Fkey` (`limit`, `limit_reset`, `limit_remaining`, `usage_daily`, `usage_monthly`). Changing limits needs the dashboard or a Management API credential (`PATCH \u002Fapi\u002Fv1\u002Fkeys\u002F{hash}`). Management credentials cannot run completions. Do not mint new secrets from an agent session.\n\nOne inference credential has **one** `limit_reset`: `daily`, `weekly`, `monthly`, or none — not daily and monthly on the same credential.\n\n### A. Per-key cap (free \u002F PAYG) — you click\n\n1. Open [https:\u002F\u002Fopenrouter.ai\u002Fsettings\u002Fkeys](https:\u002F\u002Fopenrouter.ai\u002Fsettings\u002Fkeys) (same list: [https:\u002F\u002Fopenrouter.ai\u002Fkeys](https:\u002F\u002Fopenrouter.ai\u002Fkeys)).\n2. Click the **existing** inference key. Do not recreate it. Do not paste it into Cursor.\n3. Set **Limit** to a USD credit cap for this key.\n4. Set **Limit reset** to **Daily** (midnight UTC) for a hard daily brake, **or** **Monthly** for a monthly brake — pick one.\n5. Save. Confirm with `GET \u002Fapi\u002Fv1\u002Fkey` (script `--show-limits`).\n\nTo approximate daily + monthly on PAYG without Enterprise: set the key to **daily** (hard brake) and watch `usage_monthly` from `GET \u002Fapi\u002Fv1\u002Fkey`. Monthly is then observational unless you add B.\n\n### B. Daily AND monthly hard caps together — you click\n\nWorkspace budgets can set both intervals (strictly decreasing: monthly greater than daily). Dashboard:\n\n1. Open `https:\u002F\u002Fopenrouter.ai\u002Fworkspaces\u002F\u003Cslug>\u002Fsettings` (Default workspace is allowed).\n2. Scroll to **Budgets**.\n3. Click **Add budget**, choose interval `daily`, enter a dollar amount; repeat for `monthly` (monthly must be greater than daily).\n4. Click **Save**. Organization Administrator role required. Workspace budgets are documented as Enterprise.\n\nAPI equivalent (management credential only): `PUT \u002Fapi\u002Fv1\u002Fworkspaces\u002F{id}\u002Fbudgets\u002Fdaily` and `...\u002Fmonthly` with a `limit_usd` body. Create a management credential only in the browser at [https:\u002F\u002Fopenrouter.ai\u002Fsettings\u002Fmanagement-keys](https:\u002F\u002Fopenrouter.ai\u002Fsettings\u002Fmanagement-keys).\n\nIf `GET \u002Fapi\u002Fv1\u002Fkey` shows `limit` of null, the per-key cap is not set. Treat dashboard cap as **NOT-CHECKED** until you complete A (and B if you need both windows).\n\n## Free \u002F test slugs\n\nPaid T0 (`deepseek\u002Fdeepseek-v4-flash-0731`) returned HTTP 402 on this account while slightly overdrawn. Free variants are for **sidecar smoke and cheap extract**, not agentic coding. Allowlist before the call. Catalog snapshot: [openrouter-free-models_2026-08-27.md](openrouter-free-models_2026-08-27.md).\n\nLive `GET \u002Fapi\u002Fv1\u002Fmodels` on 2026-08-27: 417 models, **20** with `pricing.prompt` and `pricing.completion` of `0` (17 `:free` chat slugs, router `openrouter\u002Ffree`, plus two Lyria audio previews that are not chat T0). Re-fetch before treating this list as current.\n\n| Role | Slug | Live result 2026-08-27 |\n| --- | --- | --- |\n| **T0 test** (smoke \u002F pong \u002F extract) | `liquid\u002Flfm-2.5-2.6b:free` | HTTP **200**. `GET \u002Fgeneration` later `total_cost=0` (provider Liquid). Liquid advises against agentic coding and may train on free traffic. |\n| **T0 test router** (random free; not for coding) | `openrouter\u002Ffree` | HTTP **200**, resolved to `nvidia\u002Fnemotron-3.5-lightning:free`. `GET \u002Fgeneration` later `total_cost=0` (Nvidia). |\n| Not a credit failure | `google\u002Fgemma-4-26b-a4b-it:free` | HTTP **429** upstream rate-limit (`Provider returned error`), not 402. |\n| **Paid cheap** (when credits exist) | `deepseek\u002Fdeepseek-v4-flash-0731` plus `:floor` | HTTP **402** Insufficient credits. Paid 3-model bake-off stays skipped until the balance is above zero. |\n\nOfficial docs say a **negative account balance may 402 even free models**. This account's live test: `:free` returned 200 while paid DeepSeek 402'd. Do not claim that is true for every overdraft; if `:free` starts 402, add credits — that is not a skill lie.\n\nPlatform caps on IDs ending in `:free`: 20 requests\u002Fminute; 50 requests\u002Fday if lifetime purchases are under $10, else 1000\u002Fday. This inference key has `is_free_tier=false` and per-key `limit=null` (dashboard Limit still a user click). `GET \u002Fapi\u002Fv1\u002Fcredits` needs `OPENROUTER_MANAGEMENT_KEY` (absent).\n\nFree vs paid training toggles are separate in [privacy settings](https:\u002F\u002Fopenrouter.ai\u002Fsettings\u002Fprivacy). Poolside `:free` pages state they may train on free inputs\u002Foutputs. Not for secrets. Do not use `:online` (deprecated) or OpenRouter web search in place of Firecrawl\u002FExa\u002Fnative Max.\n\n## Low-refusal research lane (allowlisted)\n\n`nousresearch\u002Fhermes-4-70b` — $0.13\u002F$0.40 per M, 131K ctx, reasoning-capable, RefusalBench SOTA. Lane: low-refusal\u002Fsteerable research **chat\u002Fextraction\u002Fsynthesis only** — e.g. pharmacology\u002Fharm-reduction literature queries the default hosted lanes over-refuse. Allowlist before the call, as with every slug.\n\nHARD constraints:\n\n- **No tool calling** — the OpenRouter endpoint does not expose `tools`. Never route agentic T0\u002FT1 (grunt-with-tools or T1 tool work) here.\n- **Single provider (Nebius)** — no fallback diversity; provider-sort suffixes are no-ops.\n- **Bare slug only** — no `:free` variant exists; do not append `:floor`\u002F`:nitro`\u002F`:exacto`.\n- **Paid** — 402s while the account is overdrawn; waits on the credit top-up like the other paid slugs.\n- `nousresearch\u002Fhermes-4-405b` ($1.00\u002F$3.00) is **escalation-only** for correctness-critical steerable work, per the T0→T1 escalation rules.\n\nEvidence: [openrouter-apps-hermes-uncensored_2026-08-27.md](openrouter-apps-hermes-uncensored_2026-08-27.md).\n\n## Probe commands\n\nThin T0 script: reads `OPENROUTER_API_KEY` from the environment, sends one chat completion with app headers `HTTP-Referer` and `X-OpenRouter-Title` (`agents-openrouter-sidecar`), prints generation id plus usage, then reads spend from `GET \u002Fapi\u002Fv1\u002Fgeneration?id=...` when an id is present. It never prints the secret. `GET \u002Fgeneration` can 404 for a few seconds; retry.\n\n```powershell\n# Wave 0 smoke while paid T0 is 402:\nC:\\Python314\\python.exe C:\\Users\\Prekzursil\\.agents\\scripts\\openrouter-complete.py --model liquid\u002Flfm-2.5-2.6b:free --prompt \"pong\"\n\nC:\\Python314\\python.exe C:\\Users\\Prekzursil\\.agents\\scripts\\openrouter-complete.py --show-limits --model liquid\u002Flfm-2.5-2.6b:free --prompt \"pong\"\n\n# Paid cheap (only after credits exist):\nC:\\Python314\\python.exe C:\\Users\\Prekzursil\\.agents\\scripts\\openrouter-complete.py --prompt \"Reply with the single word: pong\"\n```\n\nDo not run verbose HTTP clients against these endpoints; they can echo auth headers.\n\n## Noted, not wired (ADOPT-later)\n\n- `openrouter:web_search` server tool exists (beta) — the `:online` replacement **if ever** needed; search stays on Firecrawl\u002FMCP per routing decision.\n- Embeddings + rerank live on a **separate** catalog (`GET \u002Fapi\u002Fv1\u002Fembeddings\u002Fmodels`, 33 models, not in `\u002Fmodels`) — ADOPT-later, nothing wired.\n- `session_id` sticky routing for prompt-cache hits (~10 min stickiness) — ADOPT-later.\n\n## Related\n\n- `cost-tracking` — OpenRouter generation ids plus the Claude SQLite tracker (label the source).\n- `cost-aware-llm-pipeline` — generic routing patterns (not OpenRouter SSOT).\n- Evidence: [openrouter-complete-surface_2026-08-27.md](openrouter-complete-surface_2026-08-27.md) (38-source full-surface audit) and [openrouter-free-models_2026-08-27.md](openrouter-free-models_2026-08-27.md) (free-slug catalog snapshot).\n\n",{"name":5,"description":7},"main","a298c929b6cbd662e1fce54302ce520d5ceea1a5","skills\u002Fopenrouter-routing","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain","https:\u002F\u002Fgithub.com\u002FPrekzursil","2026-11-02 14:30:57",{"source":8,"skillName":5,"commands":49,"prompt":59,"zipAvailable":60},[50,53,56],{"tool":51,"command":52},"npx","npx skills add Prekzursil\u002Fagent-skills-toolchain --skill openrouter-routing",{"tool":54,"command":55},"bunx","bunx skills add Prekzursil\u002Fagent-skills-toolchain --skill openrouter-routing",{"tool":57,"command":58},"pnpm","pnpm dlx skills add Prekzursil\u002Fagent-skills-toolchain --skill openrouter-routing","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fopenrouter-routing\u002FSKILL.md",true,[62,68,72,78,82,86,90,94,98,102,106,110],{"path":63,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":66,"sourceHtmlUrl":67},"_tmp_hermes_endpoints.py","SCRIPT","脚本",1246,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fopenrouter-routing\u002F_tmp_hermes_endpoints.py",{"path":69,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":70,"sourceHtmlUrl":71},"_tmp_hermes_probe.py",1277,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fopenrouter-routing\u002F_tmp_hermes_probe.py",{"path":73,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":76,"sourceHtmlUrl":77},"bmad-skill-manifest.yaml","REFERENCE","参考文档",984,"https:\u002F\u002Fgithub.com\u002Fdanielmtt9\u002Fbloomcornerkiddies\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fbmad-agent-tech-writer\u002Fbmad-skill-manifest.yaml",{"path":79,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":80,"sourceHtmlUrl":81},"explain-concept.md",687,"https:\u002F\u002Fgithub.com\u002Fdanielmtt9\u002Fbloomcornerkiddies\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fbmad-agent-tech-writer\u002Fexplain-concept.md",{"path":83,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":84,"sourceHtmlUrl":85},"MAX-EFFORT-PLUGIN-CLEANUP_2026-08-27.md",6287,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fopenrouter-routing\u002FMAX-EFFORT-PLUGIN-CLEANUP_2026-08-27.md",{"path":87,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":88,"sourceHtmlUrl":89},"MAX-EFFORT-PLUGIN-DIAGNOSIS.md",15034,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fopenrouter-routing\u002FMAX-EFFORT-PLUGIN-DIAGNOSIS.md",{"path":91,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":92,"sourceHtmlUrl":93},"mermaid-gen.md",684,"https:\u002F\u002Fgithub.com\u002Fdanielmtt9\u002Fbloomcornerkiddies\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fbmad-agent-tech-writer\u002Fmermaid-gen.md",{"path":95,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":96,"sourceHtmlUrl":97},"openrouter-apps-hermes-uncensored_2026-08-27.md",24009,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fopenrouter-routing\u002Fopenrouter-apps-hermes-uncensored_2026-08-27.md",{"path":99,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":100,"sourceHtmlUrl":101},"openrouter-complete-surface_2026-08-27.md",31066,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fopenrouter-routing\u002Fopenrouter-complete-surface_2026-08-27.md",{"path":103,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":104,"sourceHtmlUrl":105},"openrouter-free-models_2026-08-27.md",11811,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fopenrouter-routing\u002Fopenrouter-free-models_2026-08-27.md",{"path":107,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":108,"sourceHtmlUrl":109},"validate-doc.md",659,"https:\u002F\u002Fgithub.com\u002Fdanielmtt9\u002Fbloomcornerkiddies\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fbmad-agent-tech-writer\u002Fvalidate-doc.md",{"path":111,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":112,"sourceHtmlUrl":113},"write-document.md",931,"https:\u002F\u002Fgithub.com\u002Fdanielmtt9\u002Fbloomcornerkiddies\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fbmad-agent-tech-writer\u002Fwrite-document.md",[115,136,167,194,208,223],{"skillNo":116,"name":117,"nameZh":6,"description":118,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":119,"categories":120,"keywords":121,"publishedAt":37},"01M6DF7D7478R55P46REYH4QAC","clarify-requirements","Resolve genuine pre-execution ambiguity by scanning repo context, generating 1-6 grounded options, and asking 1-3 focused questions via AskUserQuestion. NOT for mid-task use (injection risk); NOT when the goal is already unambiguous; use to disambiguate vague prompts before any file edits.","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Fworkspace-hooks\u002Fskills\u002Fclarify-requirements\u002FSKILL.md",[],[122,123,124,125,126,127,128,129,130,28,131,132,133,134,135],"clarify requirements","clarify","requirements","risk","options","task","prompts","goal","scanning","context","repo","injection","execution","file",{"skillNo":137,"name":138,"nameZh":6,"description":139,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":140,"categories":141,"keywords":147,"publishedAt":37},"01M6DF79CA4BHJS66J8EZDBPWJ","openmm-md","Run FREE, open-source, Windows-native GPU molecular dynamics via the openmm-md MCP (OpenMM + CUDA on the local NVIDIA GPU). Use for protein \u002F protein-ligand MD when you want a free stack or are on Windows without a Schrodinger license. Covers the full workflow (prepare, solvate, minimize, equilibrate, production, analyze, mmgbsa) AND an honest decision guide vs Schrodinger Desmond, including how to run Desmond legitimately on Linux or a cloud GPU with a real license. This is a deliberately LESSER free alternative and does NOT replace the schrodinger MCP tools. Do NOT use it to pick a free tool when a validated FEP+, WaterMap, IFD-MD, or Prime MM-GBSA result is actually required.","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Fmcp\u002Fopenmm-md-mcp\u002Fskill\u002Fopenmm-md\u002FSKILL.md",[142],{"categoryNo":143,"categoryName":144,"taxonomyType":145,"taxonomyTypeDesc":146},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[148,149,150,151,152,153,154,155,156,157,158,159,160,161,28,162,163,164,165,166],"openmm md","how","decision","pick","linux","analyze","tool","stack","open","NVIDIA","gpu","license","workflow","windows","free","molecular dynamics","local","do","MCP",{"skillNo":168,"name":169,"nameZh":6,"description":170,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":171,"categories":172,"keywords":179,"publishedAt":37},"01M6DF76C1MG85VEWP8VZCBNMN","visual-review","Take screenshots of web pages and UI using Playwright for visual review and iteration","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fvisual-review\u002FSKILL.md",[173,176],{"categoryNo":174,"categoryName":175,"taxonomyType":145,"taxonomyTypeDesc":146},"01JBSEEDCAT000000000000003","演示文稿",{"categoryNo":177,"categoryName":178,"taxonomyType":145,"taxonomyTypeDesc":146},"01JBSEEDCAT000000000000008","测试与质量",[180,181,182,183,184,185,186,187,188,189,149,190,191,192,193],"review","visual","visual review","pages","web","false","iteration","take","check","slides","ui","screenshot","playwright","screenshots",{"skillNo":195,"name":196,"nameZh":6,"description":197,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":198,"categories":199,"keywords":200,"publishedAt":37},"01M6DF73AQVCBZKEN0YXEX9FXP","status","Diagnostic status report — shows metaswarm installation state, project setup, and potential issues","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fstatus\u002FSKILL.md",[],[196,201,202,203,204,205,206,207],"setup","report","project","installation","project setup","issues","state",{"skillNo":209,"name":210,"nameZh":6,"description":211,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":212,"categories":213,"keywords":215,"publishedAt":37},"01M6DF6X6C2XB0P6GX76P7TTYV","start","Use when starting work on any task, when the user mentions metaswarm, or when the user wants to begin tracked development work","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fstart\u002FSKILL.md",[214],{"categoryNo":143,"categoryName":144,"taxonomyType":145,"taxonomyTypeDesc":146},[210,216,127,217,218,219,28,220,31,221,222],"start issue","issue","ready","work on issue","user","work on","development",{"skillNo":224,"name":201,"nameZh":6,"description":225,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":226,"categories":227,"keywords":228,"publishedAt":37},"01M6DF6QMJ9AJKFKHHKRT005PQ","Interactive project setup — detects your project, configures metaswarm, writes project-local files","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fsetup\u002FSKILL.md",[],[201,203,164,229,205,230],"files","interactive"]