Skip to main content

LLM Integration

Centy is designed from the ground up to work seamlessly with AI assistants and Large Language Models (LLMs). This page explains how to configure and optimize Centy for AI-assisted workflows.

Why AI-Native?​

Traditional issue trackers require human interaction through web interfaces. Centy's file-based approach means:

  • AI assistants can read and understand issues directly from the filesystem
  • Structured JSON metadata is machine-parseable
  • Markdown content is human-readable and LLM-friendly
  • CLI commands provide a safe interface for AI modifications

Configuration Options​

Configure AI behavior in .centy/config.json under the llm section:

{
  "llm": {
    "autoCloseOnComplete": true,
    "updateStatusOnStart": false,
    "allowDirectEdits": false
  }
}

autoCloseOnComplete​

Default: true

When enabled, AI assistants can automatically close issues after completing the associated work.

Recommended: true for automation, false for manual review workflows.

updateStatusOnStart​

Default: false

When enabled, AI assistants will automatically change issue status to "in-progress" when they begin working on it.

Recommended: true for real-time status tracking, false if you prefer manual status management.

allowDirectEdits​

Default: false

Controls whether AI assistants can directly modify files in the .centy/ folder.

Recommended: false - Always use CLI commands to ensure proper validation and manifest synchronization.

Instructions for AI Assistants​

When AI assistants encounter a Centy project, they should follow these guidelines:

DO​

  • Use centy CLI commands to manage issues and documentation
  • Read issues to understand project context and current work
  • Update issue status as work progresses
  • Close issues when work is complete (if autoCloseOnComplete is enabled)
  • Reference issue numbers in commit messages and code comments

DON'T​

  • Directly edit files in the .centy/ folder (unless allowDirectEdits is true)
  • Modify the .centy-manifest.json file manually
  • Create issue folders without using the CLI
  • Delete issues without user confirmation

Common AI Workflows​

Starting Work on an Issue​

# List open issues to find work
centy list issues --status open

# View issue details
centy show issue 1

# Update status to in-progress
centy update issue 1 --status in-progress

Completing an Issue​

# After finishing the work
centy close issue 1 --comment "Implemented feature as specified"

Creating Issues from AI Analysis​

# AI discovers a bug during code review
centy create issue "Memory leak in user session handler" --priority 2

# AI identifies refactoring opportunity
centy create issue "Refactor database connection pooling" --priority 3

Reading Project Context​

AI assistants can read the .centy/ folder to understand:

  1. Project Configuration (.centy/config.json)

    • Available states and priorities
    • Custom fields in use
    • LLM-specific settings

  2. Current Issues (.centy/issues/)

    • What work is planned
    • What's currently in progress
    • Historical context from closed issues

  3. Documentation (.centy/docs/)

    • Project documentation
    • Technical decisions
    • API documentation

Manifest​

The .centy-manifest.json file tracks project metadata:

{
  "schemaVersion": 1,
  "centyVersion": "0.1.0",
  "createdAt": "2025-01-01T00:00:00Z",
  "updatedAt": "2025-01-01T00:00:00Z"
}

AI assistants should never modify this file directly. The CLI automatically updates it when the project is initialized or modified.

Integration Examples​

With Claude​

When working with this Centy project:
1. Check open issues: centy list issues --status open
2. Update status when starting: centy update issue N --status in-progress
3. Close when done: centy close issue N --comment "reason"

With GitHub Copilot​

Centy issues can inform Copilot suggestions by providing context about current work and project requirements.

With Custom AI Agents​

Build automation that:

  1. Reads issues to determine priorities
  2. Executes work based on issue descriptions
  3. Updates status and closes issues upon completion

Best Practices​

  1. Review AI changes: Even with automation, periodically review AI-created issues and status changes

  2. Use templates: Provide templates so AI creates consistent, well-structured issues

  3. Set appropriate permissions: Use allowDirectEdits: false in production projects

  4. Monitor automation: Track AI issue closures to ensure work is actually complete

  5. Provide context: Include detailed descriptions in issues so AI understands requirements

Troubleshooting​

AI is not updating issues​

  • Check llm.updateStatusOnStart and llm.autoCloseOnComplete settings
  • Verify AI has access to run CLI commands
  • Check that issue numbers/UUIDs are correct

Manifest out of sync​

If the manifest becomes out of sync:

centy repair manifest

AI creating duplicate issues​

Implement issue search before creation:

centy search issues "login bug"

Next Steps​