aivault Runtime and Integration

This skill covers secure setup and day-to-day use of aivault for agent workflows.

Install

Install aivault if it is not already available:

curl -fsSL https://aivault.moldable.sh/install.sh | sh

Quick Reference

Default Workflow

1. •Check status and provider setup:
   • •aivault status
2. •Create provider secret (auto-provisions registry credential + capabilities when names match):
   • •aivault secrets create --name OPENAI_API_KEY --value "sk-..." --scope global
3. •Inspect available capabilities:
   • •aivault capability list
   • •aivault capability describe openai/chat-completions
4. •Invoke through policy boundary:
   • •aivault invoke openai/chat-completions --body '{"model":"gpt-5.2","messages":[{"role":"user","content":"hello"}]}'
5. •Verify audit trail:
   • •aivault audit

Detailed References

Read these as needed:

Core Usage

• •references/getting-started.md - first-run setup and secure invocation flow
• •references/cli-reference.md - command groups and practical command patterns

Security and Isolation

• •references/security-and-isolation.md - zero-trust properties, scope resolution, and common threat boundaries

Provider and Capability Design

• •references/registry-custom-providers.md - built-in registry model, schema, and custom provider flow

Operations and Debugging

• •references/operations.md - daemon behavior, environment flags, storage layout, and testing commands

Essential Patterns

1. Registry-backed secret provisioning

Use canonical secret names (for example OPENAI_API_KEY) so aivault can pin to a provider and auto-enable capabilities.

2. Capability-first invocation

Always call aivault invoke <capability-id> (or json / markdown) instead of direct upstream calls with raw keys.

3. Scoped tenancy controls

Use --scope workspace and --scope group for tenant isolation, then pass --workspace-id / --group-id on invoke.

4. Custom provider fallback

Only create manual credentials and capabilities when a provider is not in the built-in registry.

Common Mistakes to Avoid

1. •Storing API keys in .env for untrusted agent code
2. •Invoking raw URLs instead of declared capabilities
3. •Forgetting scope context when debugging credential resolution
4. •Assuming caller-provided auth headers are allowed (broker owns auth headers)
5. •Treating custom providers as equally tamper-resistant as compiled registry providers

Source of Truth

When behavior is unclear, verify against local docs and runtime artifacts:

• •~/aivault/docs/
• •~/aivault/registry/
• •~/aivault/src/