Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 

Repository files navigation

PULSE Framework

Progress-driven Unified Living System Engine

Status: Conceptual definition v0.1 — Private development
Started: June 2026


What is PULSE?

PULSE is a framework for building AI agent systems that reason about the real state of the domain to guide their own execution and evolution.

Current agent frameworks know whether they executed their steps. PULSE knows whether the work is actually done well.

The core idea

Every significant effort has dimensions that need to progress and stay healthy — not just tasks to complete. PULSE models those dimensions as Alphas: formal state indicators with verifiable checkpoints. Agents don't just execute — they evaluate whether their work has genuinely advanced those states before moving forward.

The one-sentence analogy

Like a doctor who checks your actual health before prescribing the next treatment — not just whether you followed the previous prescription.


The Problem PULSE Solves

Today's agent frameworks share a structural blind spot: they detect artifact existence, not work quality.

A workflow generates an architecture document. The system marks the phase complete. Nobody verifies whether that document actually addresses the identified risks, is coherent with requirements, or is detailed enough to build on. The project advances on an unvalidated foundation.

PULSE replaces "did the file get created?" with "do the domain checkpoints pass?"


Intellectual Foundation

PULSE synthesizes two established systems:

Essence (OMG/SEMAT)

The Object Management Group standard for software engineering methods. Contributes:

  • Alpha: Abstract-Level Progress Health Attribute — a thing whose state you need to understand to know how the project is going. Has progressive states with verifiable checkpoints.
  • to_do(): to_do(alpha, target_state) → Set<Activity> — given an Alpha's current state and a target, computes which Activities need to execute. State-driven, not sequence-driven.
  • Practice: a formally closed, composable package of Alphas + Activities + Work Products + Competencies. Practices compose algebraically.
  • Activity Space: a work container defined by Alpha-state entry and completion criteria. Closes when the world reaches the target state, not when steps finish.

BMAD (Build More, Architect Dreams)

The most adopted open-source AI agent framework (43K+ GitHub stars). Contributes:

  • Skill / SKILL.md: the universal execution unit
  • Agent types: stateless, memory, autonomous
  • Sanctum: persistent agent memory across sessions
  • Module system: installable, composable skill bundles
  • bmad-help: project navigator (extended by PULSE-Help)

PULSE takes Essence's formal state model and makes it the reasoning engine for BMAD's execution infrastructure.


Architecture

┌─────────────────────────────────────────────────┐
│                  PULSE KERNEL                   │
│  Alpha · Activity Space · Activity              │
│  Competency · Work Product · Practice           │
│  to_do() / guidance()                           │
├─────────────────────────────────────────────────┤
│              PRACTICES (extensions)             │
│  Scrum · Kanban · XP · Design Thinking          │
│  [domain-specific practices]                    │
├─────────────────────────────────────────────────┤
│           RUNTIME (BMAD-compatible)             │
│  Agent · Workflow · Skill (SKILL.md)            │
│  Sanctum + PULSE-COMPETENCY.md                  │
├─────────────────────────────────────────────────┤
│              PULSE-HELP (navigator)             │
│  Alpha evaluation · Checkpoint verification     │
│  Dynamic dependency graph · Retrograde detection│
└─────────────────────────────────────────────────┘

How PULSE Differs from Existing Frameworks

LangGraph CrewAI / AutoGen BMAD PULSE
Formal state model Partial No No Yes — Alphas
Knows work is actually done No No No Yes — checkpoints
Adapts plan to state changes No No No Yes — to_do()
Extensions derivable from models No No No Yes — Practices
Methodology included No No Yes Yes + formal
Agent memory No No Yes (Sanctum) Yes (Sanctum)
BMAD-compatible Native Yes

Proposed Alpha Kernel (hypothesis v0.1)

PULSE defines its own kernel of Alphas for AI agent systems — distinct from Essence's software engineering kernel:

Alpha States Description
Intent Identified → Clarified → Validated → Stable The goal of the effort
Context Empty → Gathered → Structured → Sufficient Information available to agents
Plan Absent → Drafted → Reviewed → Approved Work plan
Work Initiated → In Progress → Under Control → Concluded Execution state
Output Absent → Draft → Sufficient → Accepted Quality of produced results
Team Undefined → Configured → Operational → Performing Agent system state
Way of Working Undefined → Adopted → In Use → Working Well Methods and practices

These are a starting hypothesis. Defining the minimum viable kernel is one of the first open debates.


The Extension System (Practices)

Practices are formal, composable domain extensions. Unlike BMAD modules (which are written by hand), PULSE Practices are derived from formal models using PULSE-Builder — a conversational tool that transforms existing methodologies (Scrum, Kanban, XP, Design Thinking) into executable Practices:

  • Alphas specific to the methodology
  • Activity Spaces with formal entry/completion criteria
  • Activities with Approaches and required Competency levels
  • Executable SKILL.md files for each Activity

Planned Practices:

  • pulse-scrum — Sprint, Backlog Item, Sprint Goal Alphas
  • pulse-kanban — Work Item, Flow State, WIP Limit Alphas
  • pulse-xp — User Story, Test, Build Alphas
  • pulse-design-thinking — Problem Statement, Insight, Prototype, Validation Alphas

Repository Structure

pulse-framework/
│
├── README.md
│
├── docs/                        # conceptual documentation
│   ├── 01_marco_conceptual.md   # full debate — how we got here
│   ├── 02_definicion_comercial_interna.md
│   ├── 03_definicion_comercial_clientes.md
│   ├── 04_especificaciones_v01.md
│   ├── 05_resumen_contexto_proyecto.md
│   └── 06_mapa_debates_pendientes.md
│
├── kernel/                      # formal kernel definitions
│   ├── alphas/                  # Alpha definitions with states + checkpoints
│   ├── activity-spaces/         # Activity Space definitions
│   └── competencies/            # Competency definitions
│
├── practices/                   # domain extensions
│   ├── pulse-scrum/
│   ├── pulse-kanban/
│   └── pulse-xp/
│
└── runtime/                     # executable implementation
    ├── pulse-help/              # PULSE-Help navigator
    └── skills/                  # SKILL.md files for each Activity

Development Roadmap

Phase 1 — Specification (current)

  • Full conceptual framework
  • Initial specifications v0.1
  • Alpha kernel definition (minimum viable set)
  • PULSE Skill format (SKILL.md extension)
  • PULSE Alpha State File format (.pulse/state.yaml)
  • Practice composition protocol
  • PULSE-Help specification

Phase 2 — Prototype

  • PULSE-Help implementation over BMAD infrastructure
  • First Practice: PULSE-Scrum
  • Real team pilot
  • Alpha kernel iteration

Phase 3 — Ecosystem

  • PULSE-Builder: conversational Practice generator
  • Practice library for main methodologies
  • LangGraph integration as formal orchestration runtime
  • Community and Practice marketplace

Open Debates

The following design decisions are open and documented in docs/06_mapa_debates_pendientes.md:

  1. Minimum viable Alpha kernel — are the 7 proposed Alphas right? Which are redundant?
  2. Automatic checkpoint verification — LLM evaluation? Scripts? Combination?
  3. PULSE + LangGraph relationship — should to_do() be implemented over LangGraph?
  4. State retrograde handling — how does the system respond when an Alpha goes back?
  5. Boundary with validation systems — where is the line between reasoning and testing?
  6. Practice distribution mechanism — marketplace? OMG repository? Both?

License

To be determined.


Status Note

PULSE is in active conceptual development. The theoretical framework is complete. Technical specifications are in progress. The repository is currently private.


PULSE — because the pulse is the first thing you check to know if a system is alive and progressing.

About

Marco para construir sistemas de agentes de IA que razonan sobre el estado real del dominio. Los agentes actuales saben si ejecutaron sus pasos; PULSE sabe si el trabajo se hizo bien. · A framework for AI agent systems that reason about the real state of the domain, not just whether they ran their steps.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors