> ## Documentation Index
> Fetch the complete documentation index at: https://docs.lunarmc.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Workflow

> The complete lifecycle: Linear ticket → code → PR → review → merge.

# Agent Workflow

Interns operate on an automated cycle: poll Linear for work, implement changes, open PRs, and respond to review feedback.

## Linear → Code → PR

<Steps>
  <Step title="Create a Linear Ticket">
    Create an issue in [Linear](https://linear.app) on the **Mission Control** team. Tag it with the intern's label (e.g., `intern:golf`).

    The description should clearly state:

    * What needs to change
    * Which repos are affected (`layerfivecore`, `l5ui`, `docs`)
    * Acceptance criteria
  </Step>

  <Step title="Agent Picks Up the Ticket (≤5 min)">
    The `linear-poll` cron runs every 5 minutes. When it finds a new issue:

    1. Announces in Slack: `🚀 Starting MC-XXX: <title>`
    2. Registers the task in Supabase
    3. Moves the issue to **In Progress** in Linear
    4. Resets repos to the default branch
  </Step>

  <Step title="Claude Code Implements">
    The orchestrator dispatches Claude Code CLI with full tool access:

    * Creates a feature branch: `feature/mc-xxx-short-description`
    * Reads the codebase to understand patterns
    * Implements the changes
    * Commits with messages referencing the Linear issue
    * Pushes to GitHub
    * Opens PRs via `gh pr create`
  </Step>

  <Step title="Results Announced">
    After implementation:

    * PRs registered in Supabase
    * Linear issue commented with PR links
    * Slack announcement: `✅ Finished MC-XXX: PRs opened`
    * Live preview rebuilt (Angular + Django)
  </Step>
</Steps>

## PR Review Cycle

<Steps>
  <Step title="Leave a Review Comment">
    Review the PR on GitHub. Leave inline comments on specific lines or top-level review comments.

    <Tip>
      The agent detects **inline comments**, **review comments**, and **top-level PR comments** from humans. It automatically filters out bot comments (Linear, Claude, GitHub Actions).
    </Tip>
  </Step>

  <Step title="Agent Addresses Feedback (≤10 min)">
    The `pr-review` cron runs every 10 minutes. When it finds new human comments:

    1. Announces in Slack: `📝 Addressing review on l5ui#123`
    2. Dispatches Claude Code CLI to make the changes
    3. Commits: `[MC-XXX] Address review feedback`
    4. Pushes to the same branch (PR auto-updates)
    5. Replies on the PR with a summary
    6. Rebuilds the live preview
    7. Announces: `✅ Updated l5ui#123`
  </Step>

  <Step title="Merge When Ready">
    Once satisfied, merge the PR. The agent detects merged PRs and:

    * Marks the task as **completed** in Supabase
    * Frees capacity to pick up new work
  </Step>
</Steps>

## Capacity

Each intern handles **one ticket at a time**. It won't pick up new work until all PRs from the current ticket are merged or closed.

```
work-can-take → "yes" (0/1 active)
  ↓ picks up ticket
work-can-take → "no" (1/1 active)
  ↓ PRs merged
work-check-merged → frees capacity
work-can-take → "yes" (0/1 active)
```

## Cron Jobs

| Cron          | Interval     | What It Does                                           |
| ------------- | ------------ | ------------------------------------------------------ |
| `linear-poll` | Every 5 min  | Checks Linear for new tagged issues, dispatches Claude |
| `pr-review`   | Every 10 min | Checks open PRs for review comments, dispatches fixes  |

Both run as OpenClaw crons using **Claude Haiku** for orchestration (checking APIs, reading comments) and **Claude Code CLI** for actual coding work.

## Repos

The agent works across three repositories in the `Lunar-Mission-Control` org:

| Repo            | Type                 | Purpose       |
| --------------- | -------------------- | ------------- |
| `layerfivecore` | Django (Python)      | Backend API   |
| `l5ui`          | Angular (TypeScript) | Frontend app  |
| `docs`          | MDX (Mintlify)       | Documentation |

After pushing code, the agent automatically rebuilds the live preview:

* `l5ui` changes → Angular rebuild + Nginx restart
* `layerfivecore` changes → Django restart

## Example Ticket

A good ticket for an intern:

```markdown theme={null}
Title: Add console.log to login component

Description:
Add one line to the login component in l5ui.

File: src/app/components/login/login.component.ts

Add at the top of ngOnInit:
console.log("Login loaded")

One line. One file. One repo (l5ui).
```

<Warning>
  Start with small, well-defined tickets. The agent works best with clear requirements and specific file references. As you build confidence, you can assign larger features that span multiple repos.
</Warning>