LYRA 3-Brain Memory

LYRA 3-Brain Memory v2.1.0

⚠️ Persistent local storage (read before use)

This skill can write files on disk under a tree you choose with LYRA_CORE_ROOT:

• memory/YYYY-MM-DD.md — daily index
• memory/YYYY-MM-DD-.md — topic snips
• memory/reference/*.ref.txt — outer pointers
• lyra_brain_graph.json (via grow) — graph growth

Data can outlive the chat session. Do not grow secrets, tokens, private keys, or personal data you would not store on that machine.

Disable / delete: stop calling write CLIs; delete or move files under $LYRA_CORE_ROOT/memory (and graph file) yourself — this skill never auto-wipes.

Consent: write CLIs require --i-consent. Agents must not log or grow unless the user explicitly asks to remember / log / grow / recall from disk.

---

When to load this skill (narrow triggers)

Load only when the user clearly wants disk memory or 3-brain ops, for example:

• “log this session to lyra-brain” / “write a daily snip”
• “grow into the 3-brain graph”
• “recall from LYRA memory” / “brain_recall …”
• “run session_log_snip” / “3-brain memory”

Do not activate for casual “remember that…” in conversation-only context, or for “log” meaning application logging. Prefer in-chat notes unless the user asks for persistent memory.

---

Install

npx clawhub@latest install deepseekoracle/lyra-brain
export LYRA_CORE_ROOT=/absolute/path/to/LYRA_CORE   # must contain modules/lyra_brain.py

LYRA_CORE_ROOT is required. Scripts do not guess other users’ home trees without that env (or explicit flag).

Read: references/SECURITY.md · references/AGENT_CONTRACT.md · references/MEMORY_LAYOUT.md

---

Three brains (model)

---

Write commands (consent required)

Session snip

cd scripts
python session_log_snip.py --i-consent --slug moltx-hub --title "Moltx hub thread" \
  --lines "root 6073bbc0" "how map 68ebf941" --grow --ref-to MOLTX_HOW

Without --i-consent the script refuses to write.

Grow / recall

python brain_grow_cli.py --i-consent "2026-07-03 COMPLETE: lattice ALIGNED …"
python brain_recall_cli.py "Moltx how it works" --limit 5
# recall is read-only — no --i-consent required

---

Interactive runner (full power — outside skill package)

cd "$LYRA_CORE_ROOT"
python -B lyra_boot.py

---

Agent rules

1. Ask first before any disk write / grow.
2. Snips = small DONE facts (ids, urls, paths, merkle) — no essays.
3. No secrets in memory or grow text.
4. No auto git / HF / ClawHub / social publish.
5. Do not claim LATTICE ALIGNED without stack verify.
6. Do not treat vague “remember” as consent to write disk memory.

Skill chain

lygo-protocol-stack-operator → lyra-brain ↔ lyra-open-claw → lygo-ollama-army

Version

Δ9Φ963 — remember with consent and refs, not fear.