update-agents-md

Update AGENTS.md when AI causes rework, makes unauthorized decisions, or when new operational rules are needed.

Trigger Examples

• “Update AGENTS.md”
• “Add this rule”
• “Change AI behavior”
• “Don’t do X automatically”

Execution Flow

1. Read and Analyze AGENTS.md

• Load current AGENTS.md content
• Understand existing section structure
• Search for related existing rules

2. Analyze User Input

Identify the problem type:

• AI misbehavior (unauthorized decisions, rework, rule violations)
• Insufficient or ambiguous existing rules
• New operational rule addition

Determine affected section (koborin-ai specific):

• Mission
• Repository Layout
• Infrastructure Rules
• Application Rules
• Astro Development Best Practices
• CI/CD Expectations
• Release Process
• Diagram Guidelines
• Documentation Standards
• Code Quality Standards
• Pull Request Checklist

3. Determine Best Section and Placement

Extend existing section (preferred):

• Similar rules already exist in a section
• Same category

Create new section (only when necessary):

• Topic does not fit existing sections

Writing style:

• Use flat hierarchy matching surrounding items
• Write concise, factual statements
• Match existing style (English, table/list format)

4. Check Consistency with Existing Rules

• Duplicate check: Ensure same rule does not already exist
• Conflict check: Ensure no contradiction with existing rules

5. Generate and Present Update Proposal

Present in this format:

## Proposed Update

### Target Section

[Section name]

### Changes (diff format)

```diff
+ Added content

Rationale

• [Why this section was chosen]
• [Relationship to existing rules]

Approve? (yes/no)

### 6. Wait for User Approval

- **Approved**: Update AGENTS.md
- **Revision requested**: Incorporate feedback and re-propose
- **Rejected**: Cancel update

## Notes

- Match existing style: AGENTS.md uses English
- No excessive decoration: Avoid emoji or emphasis
- Concise: State facts only, avoid verbose explanations
- Manual commit: Leave commit decision to user