model-providers
Choose models and configure access via GCP Agent Platform (Vertex), Gemini API, or OpenRouter.
Install
npx skills add https://github.com/fmind/dot/tree/main/skills/model-providers
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install fmind-dot@llmmart
git clone https://github.com/fmind/dot.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole fmind/dot collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Model Providers
Configure provider auth, endpoints, and model selection. Default new integrations to GCP Agent Platform with ADC, an explicit GCP project, global, the tier's latest GA model, and high thinking (preserving OpenCode's OpenRouter default). Use API keys only on explicit request or after reporting why ADC cannot work—never auto-fallback to keys or AI Studio on error.
Workflow
- Choose the provider before checking credentials: ambient keys never dictate the provider. Read only the matching guide; gcloud owns IAM and agent-harnesses owns host configs.
- Pick tier, then verify exact GA ID: default to Gemini 4 Hydrogen (
gemini-4-hydrogen, GA 2026-11-12) for coding, agents, tool calls, and everyday execution (keepgemini-3.8-flashfor half-price high-volume extraction); escalate to frontier (Gemini 4 ArgonorCarbon; Argon GA 2026-10-13, Carbon GA 2026-11-10) for hard reasoning, architecture, or deep analysis. Look up exact IDs from the live provider catalog, never memory. - Pin backend and scope secrets: set the backend explicitly so ambient Google keys or env vars cannot hijack routing; inject keys per process via secret manager, native store, or
dot secret run NAME -- COMMAND, never in code or config files. - Probe read-only and fail closed: separate auth, model access, and bounded/timed inference checks; never silently switch providers or billing projects on failure.
Request contracts
Steer current models via instructions, thinking/effort levels, and native structured output:
- Omit sampling parameters and prefill: Gemini 3.6+ and Gemini 4 ignore
temperature/top_p/top_k(deprecation headers now; future HTTP 400) and reject prefilled trailingmodelturns with 400; Claude returns 400 on assistant prefill and sampling parameters (all models from Anthropic API2026-10-21). - Never force Claude tool calls: Claude Opus 5.5, Sonnet 5.5, and Fable 5.1+ return 400 on
tool_choiceanyortool—useautowithstrict: truetools, or structured output for JSON. - Keep prefixes byte-stable and append-only: static system text + tools first, dynamic context (date, retrieval, memory) in later turns; mutating prior turns breaks caching and invalidates thinking blocks on Claude Opus 5.5, Sonnet 5.5, and Fable 5.1+ (HTTP 400 on accounts created since 2026-08-31).
Task guides
- gcp-agent-platform: GCP Agent Platform model access with ADC and global location.
- gemini-api: Gemini Developer API access with an explicit API key backend.
- openrouter: OpenRouter model routing, API key authentication, and OpenAI-compatible requests.
Files (dot)
-
references
-
gcp-agent-platform.md 3.8 KB
Loading… -
gemini-api.md 1.7 KB
Loading… -
openrouter.md 2.5 KB
Loading…
-
-
SKILL.md 3.1 KB
--- name: model-providers description: "Choose models and configure access via GCP Agent Platform, Gemini API, or OpenRouter." license: MIT metadata: kind: collection author: Médéric HURIER (Fmind) source: github.com/fmind/dot/tree/main/skills/model-providers created: "2026-09-16" updated: "2026-11-12" --- # Model Providers Configure provider auth, endpoints, and model selection. Default new integrations to **GCP Agent Platform with ADC**, an explicit GCP project, `global`, the tier's latest GA model, and high thinking (preserving OpenCode's OpenRouter default). Use API keys only on explicit request or after reporting why ADC cannot work—never auto-fallback to keys or AI Studio on error. ## Workflow 1. **Choose the provider before checking credentials**: ambient keys never dictate the provider. Read only the matching guide; [gcloud](../gcloud/SKILL.md) owns IAM and [agent-harnesses](../agent-harnesses/SKILL.md) owns host configs. 1. **Pick tier, then verify exact GA ID**: default to **Gemini 4 Hydrogen** (`gemini-4-hydrogen`, GA 2026-11-12) for coding, agents, tool calls, and everyday execution (keep `gemini-3.8-flash` for half-price high-volume extraction); escalate to frontier (`Gemini 4 Argon` or `Carbon`; Argon GA 2026-10-13, Carbon GA 2026-11-10) for hard reasoning, architecture, or deep analysis. Look up exact IDs from the live provider catalog, never memory. 1. **Pin backend and scope secrets**: set the backend explicitly so ambient Google keys or env vars cannot hijack routing; inject keys per process via secret manager, native store, or `dot secret run NAME -- COMMAND`, never in code or config files. 1. **Probe read-only and fail closed**: separate auth, model access, and bounded/timed inference checks; never silently switch providers or billing projects on failure. ## Request contracts Steer current models via instructions, thinking/effort levels, and native structured output: - **Omit sampling parameters and prefill**: Gemini 3.6+ and Gemini 4 ignore `temperature`/`top_p`/`top_k` (deprecation headers now; future HTTP 400) and reject prefilled trailing `model` turns with 400; Claude returns 400 on assistant prefill and sampling parameters (all models from Anthropic API `2026-10-21`). - **Never force Claude tool calls**: Claude Opus 5.5, Sonnet 5.5, and Fable 5.1+ return 400 on `tool_choice` `any` or `tool`—use `auto` with `strict: true` tools, or structured output for JSON. - **Keep prefixes byte-stable and append-only**: static system text + tools first, dynamic context (date, retrieval, memory) in later turns; mutating prior turns breaks caching and invalidates thinking blocks on Claude Opus 5.5, Sonnet 5.5, and Fable 5.1+ (HTTP 400 on accounts created since 2026-08-31). ## Task guides <!-- guides:start --> - [gcp-agent-platform](references/gcp-agent-platform.md): GCP Agent Platform model access with ADC and global location. - [gemini-api](references/gemini-api.md): Gemini Developer API access with an explicit API key backend. - [openrouter](references/openrouter.md): OpenRouter model routing, API key authentication, and OpenAI-compatible requests. <!-- guides:end -->
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.