How sdi2200262/agentic-project-management works
Project Type and Purpose
– APM is an open-source framework that orchestrates a team of AI agents (Manager Agent + Implementation Agents), coordinates a persistent Memory Bank, and integrates with IDE-specific rules to manage complex projects beyond LLM context limits.
File Structure to Discover Architectural Components
• prompts/
– Contains the core prompt assets for Manager and Implementer Agents (initiation prompts, implementation-plan guides, memory-bank guides, handover protocols).
• docs/
– Holds user-facing documentation and workflow overviews (architecture explanations, integration guides).
• rules/
– Cursor IDE integration rules that “persist” scoped instructions for agents.
• assets/
– Visual assets (e.g., apm-graph).
• Memory Bank (runtime directory structure managed by the agents)
– Not explicitly in repo, but dynamically created by the Manager Agent to store logs, context, and handover artifacts.
Extract Key Architectural Roles from the README
– User/Developer: Initiates the system by launching the Manager Agent with the initiation prompt.
– Manager Agent: Orchestrator.
– Implementation Agents: Specialized workers that execute tasks.
– LLM Service: External API (e.g., OpenAI) called by all agents.
– Memory Bank: Shared, file-based persistence layer.
– Cursor Rules: Optional integration layer for enhanced reliability in a Cursor IDE.
Main Components
A. User Interface (IDE or CLI)
– Developer pastes initiation prompt or triggers automation.
B. Manager Agent
– Reads prompts/docs, creates an Implementation Plan, writes initial Memory Bank structure.
C. Implementation Agents (one or many)
– Consume task-assignment prompts, call LLM, write outputs and logs back to Memory Bank.
D. Memory Bank (Filesystem)
– Persistent storage of project logs, handover artifacts, unlogged directives.
E. External LLM Service
– Provides language-model responses to every agent request.
F. Prompt Library (prompts/ directory)
– Source of all structured prompts for agents.
G. Documentation & Assets (docs/, assets/)
– Reference material for agents and users.
H. IDE Integration Rules (rules/)
– Optional “always-on” instructions for Cursor users to reinforce context.
Relationships and Data Flows
1. Developer → Manager Agent: “Initiation Prompt”
2. Manager Agent → Prompt Library + docs/: retrieves guides
3. Manager Agent ↔ Memory Bank: initialize directories, write Implementation Plan, read/write logs
4. Manager Agent → Implementation Agents: issue “Task Assignment Prompts”
5. Implementation Agents → External LLM Service: get task output
6. Implementation Agents ↔ Memory Bank: append logs, deliver handover artifacts
7. Manager & Implementers ↔ IDE Integration Rules: (if using Cursor) continuously reinforced
Architectural Patterns and Design Principles
– Orchestration Pattern: a single Manager coordinates multiple workers.
– Separation of Concerns: prompts/docs/rules are decoupled from runtime logic.
– Pipeline & Persistence: Memory Bank acts as the stateful intermediate layer.
– Extensibility: prompt modules and rule sets can be replaced or extended for new workflows or IDEs.