TrueFoundry Agent Skills

Agent skills for TrueFoundry following the Agent Skills open format. A curated set of skills that let AI coding assistants deploy, monitor, and manage ML infrastructure.

Works with Claude Code, Cursor, Codex, OpenCode, Windsurf, Cline, and Roo Code.

Install

curl -fsSL https://raw.githubusercontent.com/truefoundry/tfy-agent-skills/main/scripts/install.sh | bash

This installs Gateway skills (the default track). To install a different track:

# Deploy skills (infrastructure, apps, jobs, etc.)
curl -fsSL https://raw.githubusercontent.com/truefoundry/tfy-agent-skills/main/scripts/install.sh | bash -s -- deploy

# All skills (gateway + deploy)
curl -fsSL https://raw.githubusercontent.com/truefoundry/tfy-agent-skills/main/scripts/install.sh | bash -s -- all

Restart your agent and start asking. If credentials are not set, your agent will prompt for them. You can also pre-set them via env vars or a .env file in your project root:

export TFY_BASE_URL=https://your-org.truefoundry.cloud
export TFY_HOST=https://your-org.truefoundry.cloud  # CLI host (same as TFY_BASE_URL)
export TFY_API_KEY=tfy-...  # https://docs.truefoundry.com/docs/generate-api-key

Do not commit .env files or API keys to Git.

If you do not have a TrueFoundry account yet, sign up first with:

uv run tfy register

tfy register is interactive. Depending on the registration server configuration, it may open a browser window for CAPTCHA or other human verification before asking you to finish email verification. After registration completes, open the tenant URL returned by the CLI, create a personal access token there, and then set TFY_API_KEY for the skills that use the platform API.

Product Tracks

• gateway installs AI Gateway-focused skills plus shared operational skills.
• deploy installs AI Deploy-focused skills plus shared operational skills.
• all installs every skill.

What You Can Do

Just ask your agent in plain English:

Gateway (default track):

• "add a new LLM provider to the gateway"
• "set up guardrails for PII filtering"
• "register an MCP server"
• "create a prompt template"

Deploy track:

• "deploy my FastAPI app"
• "launch a Jupyter notebook with a GPU"
• "deploy Postgres with Helm"

Shared (included in all tracks):

• "show logs for my-service"
• "set up a secret for my database password"
• "what's my connection status?"

Your agent picks the right skill based on what you ask. Deployment skills are explicit-only: use wording like "deploy", "helm", or "llm deploy".

Skills

Track	Skills
Gateway	ai-gateway, guardrails, mcp-servers, prompts
Deploy	applications, deploy, gitops, helm, jobs, llm-deploy, ml-repos, notebooks, service-test, ssh-server, tracing, volumes, workflows, workspaces
Shared	access-control, access-tokens, docs, logs, onboarding, secrets, status

Each skill is a standalone markdown file (skills/{name}/SKILL.md) following the Agent Skills open format.

How It Works

Skills are markdown files with instructions your agent reads at runtime. When you ask a question, your agent matches it to the right skill and follows the instructions — calling TrueFoundry APIs, running CLI commands, or both.

No SDKs to learn, no code to write. Your agent handles everything.

Development

# Edit shared files in skills/_shared/, then sync to all skills
./scripts/sync-shared.sh

# Run local validation (including offline security checks)
./scripts/validate-skills.sh
./scripts/validate-skill-security.sh

# Optional: enable pre-push hook so checks run automatically before git push
./scripts/setup-git-hooks.sh

# Install a product track and restart
./scripts/install.sh gateway
./scripts/install.sh deploy

See CONTRIBUTING.md for details on adding new skills.

Community

• Contributing Guide
• Code of Conduct
• Security Policy
• Support

License

MIT