LYGO Ollama Army & Assistant Hub

Local Ollama multi-role army + assistant hub. FULL HONEST SURFACE: (1) Default = in-process threads + STRICT name allowlist runpy (no arbitrary skill .py, no any tools/*.py). (2) Genesis localhost HTTP 127.0.0.1 only; browser open only LYGO_GENESIS_OPEN_BROWSER=1; outbound GitHub/HF/Pages probes only LYGO_GENESIS_PROBE_PUBLIC=1; no Discord/crypto/wallets in status by default. (3) Command-center: sentinel/self_tune/idle/cron/planting all gated OFF; self_tune MUTATES config when enabled and NEVER auto-enables planting. (4) Supervisor needs LYGO_ARMY_AUTONOMOUS=1 + LYGO_ARMY_I_CONSENT=1. (5) PS1 full-capacity OPERATOR-ONLY SPAWNS python.exe. (6) Social pulse / plant roles = optional consent-gated queue labels, not auto engagement. (7) No outbound webhook POST. (8) Desktop installers require LYGO_ARMY_INSTALL_DESKTOP=1. Read references/SECURITY.md first.

LYGO Ollama Army & Assistant Hub v0.8.2

SkillSpector-hardened local Ollama automation. This page is the full honest surface — not a minimal marketing blurb.

Two runtimes (do not confuse)

Runtime	Process model	Entry
Python skill surface	In-process threads + runpy allowlist	ollama_army_launcher.py, supervisor .py
Operator PowerShell	Spawns external python.exe	start_army_full_capacity.ps1 only

Manifest “no subprocess” applies to Python skill scripts. The PS1 is explicitly out of that claim.

What this skill actually does

Surface	Behavior	Default
Ollama army	Multi-role workers as in-process threads	Safe entry
Queue	Reviewed .task.json in ollama_queue/ or command_center/tasks/	Manual drop
Champions	Local persona via champion_summon.py (localhost Ollama)	Opt-in
Command center	Sentinel, self_tune, idle guardian, planting, cron	OFF until config/env
self_tune	Mutates army_config.json + may prune queue + logs	self_tune.enabled=false
Cron	Seeds safe role names (lattice/stack/pages/mesh/audit/memory) as queue tasks	Plant/social OFF
Supervisor	Long loop: sentinel + hourly cron + daemon threads	LYGO_ARMY_AUTONOMOUS=1 + LYGO_ARMY_I_CONSENT=1
Stack tools	Allowlisted in-process runpy under validated LYGO_STACK_ROOT	Opt-in
Genesis dashboard	Optional localhost HTTP 127.0.0.1:9963	Manual start
Browser open	Genesis may open system browser	LYGO_GENESIS_OPEN_BROWSER=1 only
Public probes	Optional HTTPS GET of public lattice pages	Config OFF in example
Alerts	Local logs/alerts.jsonl only	No webhook
Full-capacity PS1	Operator shell — spawns multiple python.exe	Triple env gate

Not for (defaults): remote LLM hosts, git push, HF write, ClawHub publish, autonomous social posting, silent planting, silent self_tune, silent browser open.

Leave disabled unless you need them

Flag / setting	Risk
self_tune.enabled	Config rewrite + queue prune
self_tune.auto_enable_planting	Forced false — planting never auto-on
planting.enabled + planting.consent	Kernel/registry plant roles
idle_guardian.allow_planting	Plant-like idle ops blocked unless true
idle_guardian.allow_external_memory_write	Writes into LYRA_CORE daily index
idle_guardian.allow_stack_mutating_tools	Chart rebuild / catalog render
social_publish.enabled / allow_social_pulse	Molt* pulse task seeds
access.allow_privileged_roles	egg-planter / champion-egg-boot threads
sentinel.probe_public_pages	Outbound HTTPS GET
LYGO_ARMY_AUTONOMOUS=1 + LYGO_ARMY_I_CONSENT=1	Long-running supervisor
LYGO_ARMY_FULL_CAPACITY=1 + consent + autonomous	PS1 process-spawn launcher
LYGO_GENESIS_OPEN_BROWSER=1	System browser open
start_army_full_capacity.ps1	Spawns OS Python processes

Only set LYGO_STACK_ROOT to a trusted clone.

Install

bash

npx clawhub@latest install deepseekoracle/lygo-ollama-army
ollama pull llama3.2:1b
cp ollama_command_center/config/army_config.example.json ollama_command_center/config/army_config.json

Safe first run (recommended)

bash

python ollama_army_launcher.py --model llama3.2:1b --roles hb-light,draft-simple,resonance-analyst --count 1

Autonomous supervisor (explicit dual consent)

bash

export LYGO_STACK_ROOT=/absolute/path/to/lygo-protocol-stack
export LYGO_ARMY_AUTONOMOUS=1
export LYGO_ARMY_I_CONSENT=1
python ollama_command_center/scripts/army_autonomous_supervisor.py

Operator full-capacity PS1 (process spawn — not SkillSpector Python path)

powershell

$env:LYGO_STACK_ROOT = "D:\lygo-protocol-stack"
$env:LYGO_ARMY_FULL_CAPACITY = "1"
$env:LYGO_ARMY_AUTONOMOUS = "1"
$env:LYGO_ARMY_I_CONSENT = "1"
# optional one-shots (each is another python.exe spawn):
# $env:LYGO_ARMY_RUN_SELF_TUNE = "1"   # only if self_tune.enabled in config
# $env:LYGO_ARMY_SEED_TASKS = "1"
# $env:LYGO_ARMY_RUN_CRON = "1"
.\start_army_full_capacity.ps1

Security

Read before install:

• references/SECURITY.md
• references/SECURITY_AUDIT.md
• references/SKILLSPECTOR_AUDIT.md
• references/AGENT_CONTRACT.md

Agents: propose queue JSON only; never enable planting, self_tune, autonomous, full-capacity, seed, social, or browser open without explicit user request.

Version history

Ver	Change
0.5.0	Declared permissions, webhook double-gate
0.6.0	runpy + threads; no outbound webhook
0.7.1	No auto-plant; self_tune default off; autonomous env gate
0.8.1	Dual consent supervisor; PS1 triple gate; browser open off
0.8.2	Full ClawHub audit: strict runpy allowlist; local-only genesis collect; no bak planting config; health probes-only default; no cross-skill cron; no Discord/crypto status; no autostart/push commands; desktop installers gated

Δ9Φ963 — local flame, reviewed queue, allowlisted tools, no silent outbound, honest agency.