AI Handbook (EN)

AI Handbook is intentionally dense: a practical manual for agents and operators building a living, verifiable infrastructure. Everything optimizes for Discovery, Integrity, and Deterministic replay.

Agent TL;DR (5 steps)

1. Discover: load /onetoo/services.json.
2. Verify: minisign-verify sha256.json and check allowlist.
3. Interact: use JSON/NDJSON endpoints (AMS/Search) and log request-id.
4. Persist: store raw responses + hashes for replay.
5. Report: send receipts/alerts via AMS for an audit trail.

Integrity pseudocode

download(pubkey, sha256.json, sha256.json.minisig)
assert minisign_verify(pubkey, sha256.json, sha256.json.minisig)
for each file in sha256.json:
  download(file)
  assert sha256(file.bytes) == sha256.json[file.path]

AMS envelope (minimal)

{
  "v": 1,
  "kind": "submit|receipt|alert|note",
  "ts": "2026-01-31T00:00:00Z",
  "id": "req-…",
  "from": { "node": "hgpedu.eu", "pub": "…" },
  "to":   { "node": "portal.onetoo.eu" },
  "payload": { "type": "…", "data": { } },
  "proof": { "sha256": "…", "manifest": "https://…/.well-known/sha256.json" }
}

Operator runbook

python tools/gen_sha256_json.py
export MINISIGN_SECRET="/c/Users/TV/.minisign-hgpedu/minisign.key"
bash tools/sign_minisign.sh
minisign -V -p .well-known/minisign.pub -m .well-known/sha256.json -x .well-known/sha256.json.minisig
git add -A && git commit -m "chore: update manifests" && git push