NeuroLift TOI Framework (Solidarity)

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"

Terms of Interaction (.toi) is the NeuroLift standard for user-controlled, neurodivergent-friendly, privacy-first AI collaboration — a small, portable, human-editable JSON file that states how AI systems should interact with a person.

This repository is the TOI layer of the Solidarity Framework. It ships two reference implementations of the same .toi v1.0.0 standard — one in Python (nlt-toi) and one in TypeScript (@neurolift-technologies/toi) — plus the human-facing templates, governance records, and runtime experiments.

The Python library is a faithful port of the TypeScript reference: the on-disk format, RFC 8785 canonicalization, and Ed25519 signature envelope are identical, so a document signed by one implementation verifies in the other.

Repository orientation

1. .toi Python reference library — nlt_toi/ (src/nlt_toi/): parse, validate, RFC 8785 canonicalize, tier-precedence resolve, and Ed25519 sign/verify. Published to PyPI as nlt-toi.
2. .toi TypeScript reference library — packages/toi/: the normative reference, the 1.0.0 specification (SPEC.md), the JSON Schema artifact, and the conformance fixtures. Published to npm as @neurolift-technologies/toi.
3. TOI-governed agent experiments — Python and GitHub Pages examples in src/fusion/ and docs/ demonstrating TOI as runtime context for an assistant.

The governance files at the repository root (AGENTS.md, NLT-DEV-OTOI.md, nltotoi.json, and docs/active-threads.md) are part of the working system, not incidental documentation. Coding agents must follow them before changing code or docs.

What is included

Quick start: nlt-toi (Python)

pip install nlt-toi

from nlt_toi import (
    parse_toi, safe_parse_toi, serialize_toi,
    canonicalize, generate_key_pair, sign_toi, verify_toi,
    resolve_toi,
)

# Parse + validate (accepts a JSON string or an already-parsed dict).
doc = parse_toi('''
{
  "$toi": "1.0.0",
  "$tier": "personal",
  "identity": { "author": "alice" },
  "communication": { "tone": "direct", "verbosity": "concise" }
}
''')

# Sign and verify. The Ed25519 signature is detached over the canonical payload
# (the document with $signature removed) and carried back in the $signature key.
keys = generate_key_pair()
signed = sign_toi(doc, keys.private_key)
assert verify_toi(signed) is True

# RFC 8785 canonical form — the exact bytes that get signed.
print(canonicalize(doc))

# Resolve precedence: personal > community > project > platform defaults.
project_doc = parse_toi('{"$toi":"1.0.0","$tier":"project","identity":{"author":"acme"}}')
community_doc = parse_toi('{"$toi":"1.0.0","$tier":"community","identity":{"author":"team"}}')
personal_doc = doc  # the personal-tier document above
effective = resolve_toi([project_doc, community_doc, personal_doc])

Non-throwing parsing is available via safe_parse_toi, which returns a SafeParseResult (.success, .data, .error) instead of raising.

The public API mirrors @neurolift-technologies/toi (in snake_case): parse_toi/safe_parse_toi/is_toi/serialize_toi, canonicalize/ canonicalize_to_bytes, generate_key_pair/sign_toi/verify_toi/is_signed/ signing_payload, and resolve_toi/sort_by_precedence/compare_tier.

Quick start: @neurolift-technologies/toi (TypeScript)

cd packages/toi
npm install && npm test && npm run build

Primary references for the standard:

• Specification: packages/toi/SPEC.md
• Generated JSON Schema: packages/toi/schema/toi-1.0.0.schema.json
• Source exports: packages/toi/src/index.ts
• Conformance fixtures: packages/toi/test/fixtures/ (the Python library is tested against this same fixture set)

TOI-governed agent demo

The GitHub Pages demo in docs/ shows one possible TOI-governed assistant UI: it injects a TOI JSON document as system context for a prompt. The demo code is intentionally small and provider-specific so it is easy to inspect. It is not a production deployment pattern. External service choices, production deployment, and architecture decisions require explicit approval under NLT-DEV-OTOI.md. Personal tokens and TOI data must never be committed to this repository.

Documentation map

• Framework definition: docs/framework-overview.md
• Developer implementation guide: docs/implementation-guide.md
• Best practices: docs/best-practices.md
• NeuroLift multi-agent integration playbook: docs/neurolift-integration.md
• Development process and CI runbooks: docs/development-process.md

Development and validation

# Python .toi library tests (uses the npm conformance fixtures)
pytest

# TypeScript reference tests
cd packages/toi && npm test && npm run build

# Governance file validation
bash .nltotoi/scripts/validate-governance.sh

Notes:

• The Python library depends on jsonschema (validation) and cryptography (Ed25519). It targets Python 3.9+.
• The governance validation script reports missing or stale required governance files; resolve failures before merging governance-related changes.
• The TypeScript package requires Node.js 18 or newer.

Contributing

Start with CONTRIBUTING.md, CODE_OF_CONDUCT.md, and SECURITY.md. Coding agents must additionally read, in order: NLT-DEV-OTOI.md, CLAUDE.md, docs/active-threads.md, then self-register and leave a handoff record for significant sessions.

License

Apache-2.0 (see LICENSE). The @neurolift-technologies/toi package also ships its own Apache-2.0 license copy at packages/toi/LICENSE.