Skip to content

AGENT_RESUME.md

Latest commit

 

History

History
81 lines (55 loc) · 2.51 KB

AGENT_RESUME.md

File metadata and controls

81 lines (55 loc) · 2.51 KB
Raw

Perpetua-Tools — Agent Resume

Hardware Safety & Model Affinity (2026-04-26)

Canonical policy: config/model_hardware_policy.yml

Hard rules — never override:

  • Windows-only (NEVER_MAC): gemma-4-26b-a4b-it, gemma-4-26B-A4B-it-Q4_K_M, qwen3.5-27b-claude-4.6-opus-reasoning-distilled-v2
  • Mac-only (NEVER_WIN): gemma-4-e4b-it, qwen3.5-9b-mlx, qwen3.5-9b-mlx-4bit

Runtime validation fires at three layers (L1 discover, L2 manager/launcher, L3 API). Any HardwareAffinityError must escalate to Controller — never silently fallback to unsafe hardware.

If you are an AI agent reading this: do NOT add unverified model IDs to any policy file or config. Confirm with discover.py --status on actual hardware first. Known hallucinations removed: qwen3-coder-14b and gemma4:e4b.

Status: COMPLETE ✅ (2026-04-20)

Branch: main

All Claude Code automation from the LM Studio Auto-Discovery plan has been implemented.

What Was Done

Config Updates

File Change Status
config/devices.yml mac-studio lan_ip .103→.107, win-rtx3080 .100→.101 ✅
config/models.yml LM_STUDIO_MAC_ENDPOINT default → localhost, WIN → .101 ✅
.gitignore .env.lmstudio added ✅

.claude/ Automations

Type File Status
Hook: SessionStart .claude/settings.json ✅ discover-lm-studio.sh (async) + sync-companion-instincts
Hook: PostToolUse .claude/settings.json ✅ ruff check on .py + pytest tests/
Skill .claude/skills/agent-run/SKILL.md ✅ env validation before agent launch
Skill .claude/skills/model-routing-check/SKILL.md ✅ Claude-only endpoint check
Subagent .claude/agents/api-validator.md ✅ Perplexity + LM Studio schema validation

Shell Gate

scripts/discover-lm-studio.sh — Layer B gossip gate (5-min TTL, delegates to ~/.openclaw/scripts/discover.py)

Live Endpoints (auto-updated by discover.py)

Mac LM Studio: http://localhost:1234
Win LM Studio: http://<YOUR_WINDOWS_IP>:1234
LMS_WIN_MODEL: qwen3.5-27b-claude-4.6-opus-reasoning-distilled-v2
LMS_WIN_FALLBACK_MODEL: gemma-4-26b-a4b-it

How to Resume

# Check LM Studio routing before running agents
/model-routing-check

# Launch orchestrator with env validation
/agent-run

# Validate API response schemas
# (subagent: api-validator)

# Force-refresh endpoints
python3 ~/.openclaw/scripts/discover.py --force

# Check status
python3 ~/.openclaw/scripts/discover.py --status