ASFDK — Agent Solidarity Framework Development Kit

ai_assistant_directive:
  priority: 1
  instruction: "Before performing any development or coding tasks, read NLT-DEV-OTOI.md."
  description: "NLT-DEV-OTOI.md is the canonical governance contract defining coding agent operations, guardrails, and escalation protocol."
  required_file: "NLT-DEV-OTOI.md"

NeuroLift Technologies' reference implementation of the HAIEF Solidarity Framework.

One kit, two runtimes: TypeScript (npm) and Python (PyPI).

ASFDK is governance and protection middleware that sits between the model and the agent runtime. It implements the Solidarity Framework standard maintained by HAIEF and packages it with a high-level orchestrator (NeuroLiftFoundation) and re-exports of all four pillar packages.

ASFDK gives you a single install that surfaces every layer of the model ↔ agent boundary:

[Model Provider]
      ↓
[ASFDK Solidarity Layer]
   • User-preference governance (TOI)
   • Multi-agent coordination (OTOI)
   • Crisis detection & response (RRT Advocate)
   • Continuity across drift (Sleepwalker Protocol)
      ↓
[Agent Runtime or Claws (Agent Wrappers)]
      ↓
[Tools, APIs, Actions]

Pick your runtime:

• TypeScript/Node.js → npm install @neurolift-technologies/asfdk (primary, published from packages/asfdk/)
• Python → pip install asfdk (faithful port, published from src/asfdk/)

Note on “claws” (Agent Wrappers): In ASFDK documentation, claws refers to wrapper implementations that orchestrate model calls, tools, and runtime behavior.

Install

TypeScript (npm) — Primary

npm install @neurolift-technologies/asfdk

The four pillar packages are declared as dependencies, so they are installed transitively:

Python (PyPI) — Faithful Port

pip install asfdk

The Python package mirrors the TypeScript API and depends on the Python equivalents of the pillars:

Quick Start (Both Runtimes)

TypeScript

import { createFoundation, FoundationMode, InteractionType, toi } from '@neurolift-technologies/asfdk';

async function main() {
  // Orchestrator: route interactions through the active components for a mode.
  const foundation = await createFoundation('user-123', FoundationMode.UNIFIED);

  const response = await foundation.processInteraction({
    timestamp: new Date(),
    interactionType: InteractionType.PREFERENCE_UPDATE,
    data: { toi: { $toi: '1.0.0', $tier: 'personal', identity: { author: 'user-123' } } },
    userId: 'user-123',
  });

  // Pillars are also available directly as namespaces.
  const parsed = toi.safeParseToi(myPreferences);
}

main();

Python

import asyncio
from asfdk import create_foundation, FoundationMode, InteractionType, toi

async def main():
    foundation = await create_foundation("user-123", FoundationMode.UNIFIED)

    response = await foundation.process_interaction({
        "timestamp": datetime.now(timezone.utc),
        "interaction_type": InteractionType.PREFERENCE_UPDATE,
        "data": {"toi": {"$toi": "1.0.0", "$tier": "personal", "identity": {"author": "user-123"}}},
        "user_id": "user-123",
    })

    # Pillars are also available directly as namespaces.
    result = toi.safe_parse_toi(my_preferences)

asyncio.run(main())

Foundation Modes

FoundationMode controls which Solidarity Framework components are active at runtime.

Per-component overrides are available via FoundationConfig.components.

Rollout Phases (Recommended Practice)

When integrating ASFDK into an existing system, work through these phases. They are operator-applied via component config and thresholds — not separate runtime modes.

1. Observe — Deploy with high thresholds so the layer logs decisions but rarely intervenes. Use logs to calibrate.
2. Advise — Lower thresholds gradually; emit warnings to the agent/operator but don't gate model output yet.
3. Enforce — Apply governance decisions inline. Promote to production only after an nlt-redteam review pass.

⚠️ Crisis detection is a prototype — not a safety system

The RRT Advocate layer wraps an experimental crisis-detection library with stubbed intervention layers. It is not medical advice, not a crisis service, performs no real-time monitoring, and can miss real crisis signals. Never rely on it as the sole safety mechanism. If you or someone else needs help now, in the US call or text 988 or chat 988lifeline.org.

Companion Tools

nlt-toi is a pre-flight tool, not an ASFDK component. The ASFDK enforces TOI at runtime via the NLT-OTOI framework. Use nlt-toi upstream to generate well-formed, validated TOI documents that the OTOI layer can consume.

Repository Structure

asfdk/
├── packages/asfdk/          # npm @neurolift-technologies/asfdk (TypeScript source)
├── src/asfdk/               # PyPI asfdk (Python source, faithful port)
├── hosting/                 # Next.js landing page (separate deliverable)
├── workers/                 # Cloudflare Workers deployment (separate deliverable)
├── legacy/                  # Archived Python-era implementation (v0.1.x)
│   ├── unified_core/        # Old ASFDK Python core
│   ├── nlt-otoi/            # Vendored NLT-OTOI (superseded by npm @neurolift-technologies/otoi)
│   ├── rrt-advocate/        # Vendored RRT Advocate (superseded by npm @neurolift-technologies/rrt-advocate)
│   ├── sleepwalker/         # Vendored Sleepwalker (superseded by npm @neurolift-technologies/sleepwalker-protocol)
│   ├── config/              # Python-era config
│   ├── scripts/             # Python sync scripts
│   ├── tests/               # Python-era tests
│   ├── DEPRECATED.md        # Archive index and migration guide
│   └── ...                  # Dockerfile, pyproject.toml.v0.1.0, requirements.txt, GEMINI_TOPOGRAPHY.py
├── docs/                    # Documentation (dev/deploy quickstarts, etc.)
├── .nltotoi/                # NLT governance namespace
├── NLT-DEV-OTOI.md          # Org-level coding agent contract
├── AGENTS.md                # Repo-specific agent guidance
├── CLAUDE.md                # Project-specific context
└── LICENSE                  # Apache-2.0

License

This project is licensed under the Apache License 2.0 — see the LICENSE file for details.

Acknowledgments

• NeuroLift Technologies — Core AI-fusion framework and methodology
• Human & AI ElevAItion Foundation (HAIEF) — Governance standards
• ADHD Community — Feedback and real-world testing
• Open Source Contributors — Various libraries and tools