Agent Skills
Connect a customer project to TokGate with a coding agent, an API key stored securely, an exact model ID and a clear implementation request.
On this page
What a TokGate Skill does
A Skill is an instruction package for coding agents such as Codex. It tells the agent how to select the correct TokGate protocol, modify the customer project, keep credentials server-side and verify the result. It is not an SDK and does not call a model merely by being installed.
Install the Skills
Run the command in the customer project. Project installation keeps the Skill with that project; add --global only when the same user should access it from every project.
npx skills add tokgate/skillsInstall one focused Skill by adding --skill. This example installs the chat workflow for Codex.
npx skills add tokgate/skills --skill tokgate-chat --agent codex --yesChoose the right Skill
Use the focused Skill when the task is limited to one capability. Use the integration Skill for initial setup, migrations or work spanning several API families.
| Skill | Use it for |
|---|---|
tokgate-api-integration | Initial setup, Base URL configuration, protocol selection, migrations and gateway-wide troubleshooting. |
tokgate-chat | Chat completions, streaming, tool use, structured output and vendor-native chat formats. |
tokgate-image | Image generation and editing with the documented Gemini, GPT Image, Seedream or Qwen Image family. |
tokgate-video | Seedance or HappyHorse task submission, polling, terminal-state handling and output download. |
Give the agent four pieces of information
The shortest reliable request contains the project, the installed Skill, the exact model ID and the required behavior. Configure the credential before asking the agent to implement the integration.
- Project: open the repository that should be changed and state its runtime or framework when it is not obvious.
- Skill: name the focused Skill explicitly, for example
$tokgate-chat. - Model ID: copy the exact ID from the Model Catalog or query
GET /v1/models. Do not derive it from a display name. - Outcome: describe streaming, tools, image editing, polling, framework constraints and the verification level you need.
Example customer request
The API key is not part of the prompt. The customer configures it in the environment or a secret manager and tells the agent only where it is available.
Use $tokgate-chat to connect this Node.js project to TokGate.
Model ID: YOUR_MODEL_ID
The API key is already stored in TOKGATE_API_KEY.
Keep it on the server, add streaming, and verify with mocks before any billable request.Keep the API key out of the conversation and browser
The customer should configure TOKGATE_API_KEY in a local environment, deployment Secret or server-side secret manager. Do not paste its value into the prompt.
- Do not commit credentials to Git, examples, screenshots or logs.
- A browser application must call its own backend; the backend calls TokGate.
- Allow the agent to read the environment variable only when a real verification request is required. It should never print the value.
Installation does not spend CreditsListing, installing and reading a Skill are local operations. A real chat, image or video generation request can consume Credits, so ask the agent to use mocks or static checks first and authorize a billable request explicitly.
What the agent should verify
A completed integration should identify the selected protocol and endpoint, list changed files, run the project checks, and provide a credential-free verification command. For video, it must preserve the task ID and poll the matching status endpoint instead of resubmitting.
- Before a live call: validate request construction with fixtures, mocks, types or static checks.
- For a live call: confirm the exact model is callable for the active key and obtain customer authorization for possible Credit usage.
- After implementation: report the protocol, endpoint, environment variable, changed files and test result without exposing credentials or signed media URLs.