Chat mode imported from JohnMichaelMiller/ai-practitioner-resources (.github/chatmodes/PM_MODES_OVERVIEW.chatmode.md). Copyright stays with the author.

Project Manager Modes Overview

This document describes the two PM modes available in the AI Practitioner Resources project and explains their relationship.

Two PM Modes with Equivalent Intent

Both modes serve the same purpose: triaging, prioritizing, and managing issues in a GitHub repository. They share the same core responsibilities but target different usage contexts.

Mode 1: Project Manager (Chat Mode)

File: .github/prompts/modes/project-manager.md

Context: Long-form chat interactions with human project managers

Use case:

• Manual issue review and prioritization sessions
• Interactive lane rebalancing discussions
• Deep-dive issue analysis with back-and-forth Q&A
• Training and documentation reference for human PMs

Characteristics:

• Comprehensive guidance with detailed examples
• Includes PowerShell and CLI command examples
• Extensive edge case handling
• Approval protocol and review checklists
• Full scoring rubric with tie-breaking rules

Mode 2: AI Assistant PM

File: .github/prompts/modes/ai-assistant-pm.md

Context: Automated GitHub Actions workflow execution

Use case:

• Automated issue intake on issue open events
• Rapid initial triage with structured JSON output
• Token-optimized for cost efficiency
• Batch processing multiple issues

Characteristics:

• Strict JSON-first output protocol
• Concise rationales (2-4 sentences max)
• Token optimization guidelines
• Batch operation formats
• Focus on speed and automation

Binding Relationship

Both modes implement the same decision framework but with different presentation styles:

Shared Core Principles

1. Lane Management

   • Four swimlanes: at bat (3 max), on deck (3 max), in the hole (3 max), on the bench (unlimited)
   • Every issue has exactly one lane label
   • Rebalance only on close events, not on open/edit

2. Approval Gate

   • Only issues with "implementation ready" label (applied by humans) can enter active pipeline
   • Unapproved issues stay on the bench

3. Independence Rules

   • Active lane issues must be implementable simultaneously
   • No merge conflicts or overlapping file changes
   • Different components/folders preferred

4. Priority Scoring

   Priority Score = Impact (0-5) + Urgency (0-5) + (5 - Risk) + (5 - Size penalty) + Independence (0-5)
   

   • Tie-breaking: (1) independence > (2) smaller size > (3) older issue

5. Required Labels

   • Type: feature, bug, refactor, documentation, etc.
   • Size: small, medium, large
   • Priority: priority:NN (0-100)
   • Independence: independence:high or independence:low
   • Risk: risk:low, risk:medium, risk:high
   • Readiness: needs-clarification or implementation ready

6. Size Constraints

   • size:large issues MUST be split into smaller sub-issues
   • Only size:small and size:medium can be approved

Key Differences

When to Use Each Mode

Use Project Manager Mode When:

• Manually triaging issues in a chat session
• Training new team members on the PM process
• Discussing edge cases or complex dependencies
• Conducting quarterly/monthly prioritization reviews
• Resolving conflicts between competing priorities

Use AI Assistant PM Mode When:

• Automating issue intake via GitHub Actions
• Processing high volumes of issues quickly
• Requiring consistent, structured output for parsing
• Minimizing API costs through token optimization
• Integrating with downstream automation tools

Synchronization

Both modes reference the same source of truth:

• Lane definitions and caps are identical
• Approval criteria match exactly
• Scoring formulas produce the same results
• Label naming conventions are consistent

Changes to rules or criteria should be propagated to both files to maintain equivalence.

Migration Between Modes

If you need to switch contexts:

From Chat to Automation:

• Extract the structured decision (JSON) from the conversation
• Apply labels and project status programmatically via gh CLI or API

From Automation to Chat:

• Review the JSON output from the automation
• Use the human-readable review comment as context
• Continue discussion based on the automated assessment

Implementation Details

Workflow Integration

1. Issue Intake (Automated)

   • Workflow: .github/workflows/issue-intake.yml
   • Script: scripts/issue-intake.js (project board setup)
   • Script: scripts/pm-review.js (AI assistant PM mode)
   • Uses AI Assistant PM mode for initial triage

2. Lane Rebalancing (Automated)

   • Workflow: .github/workflows/rebalance-on-close.yml
   • Script: scripts/rebalance-lanes.js
   • Uses scoring logic consistent with both modes

3. Manual Review (Chat)

   • Script: scripts/pm-review-local.js
   • Uses Project Manager mode for interactive sessions

Best Practices

1. Consistency First: Always apply the same criteria regardless of mode
2. Document Changes: Update both mode files when rules change
3. Validate Equivalence: Test that both modes produce similar decisions for the same issue
4. Token Budget: Use AI Assistant mode for automation; save Project Manager mode for human interactions
5. Audit Trail: Both modes should reference the issue in comments for traceability

See Also

• Project Manager Mode Details
• AI Assistant PM Mode Details
• PM Review Prompt
• Issue Intake Script
• Rebalance Lanes Script