Introduction

CODI’s modularity stems from its built-in agents. Each agent encapsulates a single responsibility, enabling predictable orchestration and straightforward customization. This page details the purpose, inputs, outputs, and extension points for each core agent.

1. Scaffold Agent (scaffold-agent)

• Purpose: Bootstraps a new Anchor project with directory structure, boilerplate code, and configuration files.

• Inputs: --template flag (e.g., anchor, spl-token, custom templates).

• Outputs: programs/<name>/lib.rs, Anchor.toml, frontend TypeScript stubs, sample tests.

• Hooks: Pre/post hooks in codi.toml:

  [scaffold]
  
  

2. Audit Agent (audit-agent)

• Purpose: Runs static and semantic security analyses on program code.

• Inputs: Source files from scaffold or user workspace.

• Outputs: JSON/Markdown audit report with severity levels, remediation suggestions, and metadata.

• Extension: Integrate custom linters by specifying:

  [audit]
  linters = ["slither", "custom-lint"]
  
  

3. Test Agent (test-agent)

• Purpose: Auto-generates and executes unit and integration tests based on code and comments.

• Inputs: Program IDL, Rust comments, developer annotations (e.g., /// @test-case).

• Outputs: Test suites (Mocha/Chai or Rust), test coverage metrics, console reports.

• Customization: Use @codi: annotations in Rust to guide test generation:

  /// @codi:test description="Ensure transfer fails on insufficient funds"
  pub fn transfer(ctx: Context<Transfer>
  
  

4. Deploy Agent (deploy-agent)

• Purpose: Compiles, packages, and deploys the Anchor program to the specified Solana cluster.

• Inputs: Compiled artifacts, network flag (--network devnet|testnet|mainnet).

• Outputs: Program ID, transaction signature, IDL file on chain.

• Rollbacks: Enabled by default; snapshots last successful deploy in mcp.deploy node.

• Config: Customize via codi.toml:

  [deploy]
  
  

Visual Description

1. Agents Sequence Diagram:

   • Timeline showing scaffold → audit → test → deploy.

   • Data passed between agents via MCP context nodes.

2. Configuration Schema Snippet:

   • YAML-like depiction of codi.toml sections for all four agents.

Conclusion

CODI’s built-in agents encapsulate each stage of the smart-contract lifecycle, offering clear extension points and configuration options. Customize or disable any agent to tailor CODI to your team’s workflow.