Skip to main content

Agent Skills

Use the Vluna skill bundle when you want a coding agent to implement Vluna integration in your codebase with a predictable workflow.

The bundle is designed for customer repositories where the agent needs:

  • the core Vluna concepts and guardrails
  • offline reference documents for SDK integration
  • a repeatable order of work for authorize -> commit/cancel

What the skill is for​

The Vluna skill helps a coding agent:

  1. collect the missing integration inputs first
  2. choose the right runtime boundary
  3. implement runtime metering before broader billing scope
  4. use the bundled references when offline
  5. use hosted docs as supplements when network access is allowed

This is most useful when the repository already contains one of these runtime shapes:

  • direct LLM SDK calls
  • OpenAI Agents SDK
  • Google ADK
  • LangGraph or LangChain
  • a backend service that needs direct Vluna SDK or HTTP integration

Download the bundle​

Download the latest hosted bundle here:

The zip expands to:

vluna/
  SKILL.md
  LICENSE.txt
  references/
  scripts/

What each bundle file is for​

  • SKILL.md
    • Main instructions for the coding agent
  • references/
    • Offline background, integration guides, endpoint index, troubleshooting, and LLM-specific guidance
  • scripts/repo_intake.sh
    • Optional helper for quickly scanning an unfamiliar repository
  • LICENSE.txt
    • Bundle license terms

Install locations​

The exact location depends on the coding agent.

Claude Code​

Claude Code supports native Skills discovery.

Place the extracted vluna/ directory in one of these locations:

  • project-local: .claude/skills/vluna/
  • user-level: ~/.claude/skills/vluna/

Sanity check:

  • ask What Skills are available?

Other coding agents​

For tools without native Skills discovery, keep the extracted bundle in the repository and explicitly tell the agent to read it.

Simple pattern:

  • place vluna/ somewhere stable in the repo
  • in the first prompt, tell the agent to read vluna/SKILL.md before making changes

Use this flow regardless of the coding tool:

  1. extract the bundle into the repository or agent skills directory
  2. start the agent in the repository root
  3. tell the agent to read SKILL.md first
  4. require it to ask for missing high-risk integration inputs before editing
  5. have it implement one real authorize -> commit/cancel path first
  6. only then expand into additional models, tools, or outcome billing

Suggested first prompt​

Use the Vluna skill. Read vluna/SKILL.md first. Before editing anything, ask me for the missing integration details required by the skill. Then implement one real authorize -> commit/cancel integration path with minimal changes, run the smallest relevant verification, and summarize the patch plus the configuration I must carry to production.

What the skill expects from the developer​

Be ready to provide:

  • deployment type: Cloud or self-hosted
  • realm_id
  • how principal_id should be defined in your system
  • whether billing_account_id is also needed in your integration
  • where the correct runtime boundary lives
  • whether you already have a feature or meter catalog
  • whether local tests or dev commands may be run by the agent

Offline and online references​

The bundle is designed to work offline first.

  • Use the bundled references/ directory as the primary source when the agent has no network access.
  • Use hosted docs only as supplements when online access is available.

The most useful hosted supplements are listed inside SKILL.md.

Skill vs MCP​

Use the skill when you want the agent to modify code in a customer repository.

Use the MCP server when you want the agent to inspect or operate Vluna resources through tools at runtime.

They are complementary:

  • skill: implementation playbook
  • MCP: runtime API and operations access

Troubleshooting​

The agent ignores the skill​

  • explicitly instruct it to read SKILL.md first
  • repeat that it should ask for missing integration details before editing

The agent starts coding too early​

Put this at the top of your first message:

  • Before editing anything, ask me for the missing high-risk integration details listed in SKILL.md.

The agent cannot access hosted docs​

  • continue with the bundle's offline references
  • the bundle is expected to remain usable without network access

The agent does not know which guide to use​

Point it to:

  • references/integration-guide.md for the master flow
  • references/llm-integration-guide.md for AI and agent products
  • references/runtime-metering-checklist.md for the shortest path