菜单
Run the local NullClaw host security app for skill governance, curated-source discovery, guarded threat suppression, and self-governance on Windows hosts.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Sweep one or more local repositories for implementation gaps and produce deterministic, evidence-backed remediation lanes. Use when you need cross-repo detection of missing tests, docs lockstep drift, risky TODO/FIXME additions, and release-hygiene misses.
Build and query bounded metadata-only repository indexes without rg process churn. Use when working in very large repos and you need deterministic file discovery by path, extension, language, scope, or freshness.
Detect recent cross-repo work and produce deterministic JSON evidence for skill curation and routing decisions. Use when workflows span multiple repos and you need a bounded view of recent commits, touched files, and policy/contract-sensitive changes.
Enforce hardware-first diagnosis and fixes in <PRIVATE_REPO_B>. Use for runtime probe, telemetry, inference, and integration failures where strict real-hardware behavior, no new stubs, deterministic diagnostics, and no unrequested driver/runtime mutation are required.
Run a defender-first security sweep on code, configs, prompts, model/tooling surfaces, or third-party contribution lanes. Use when a request involves safe bug, leak, zero-day-class, exploit, or hack hunting for protection, when contributing to outside repositories and you want a focused security pass, or when touching auth, secrets, permissions, network exposure, prompt/tool boundaries, data flow, or update/build surfaces. This skill is defensive only and must never be used for weaponization or unauthorized access.
Present, navigate, snapshot, and hide HTML content on connected OpenClaw nodes (Mac, iOS, Android) via the canvas tool. Use when displaying games, visualizations, dashboards, or interactive demos on remote node canvas views, or when debugging canvas URL routing across loopback, LAN, and Tailscale bind modes.
| name | skill-arbiter |
| author | grtninja |
| canonical_source | https://github.com/grtninja/skill-arbiter |
| description | Run the local NullClaw host security app for skill governance, curated-source discovery, guarded threat suppression, and self-governance on Windows hosts. |
Use this skill when the work requires:
Install desktop launcher dependencies when using the managed security-console startup flow:
cd apps/nullclaw-desktop
npm install
Open the desktop app for developer/debug work:
python scripts/nullclaw_desktop.py
Open the desktop app through the accepted no-shell Windows launcher:
wscript.exe //B //Nologo .\scripts\launch_security_console.vbs
Run the local loopback agent without the desktop shell:
python scripts/nullclaw_agent.py
Refresh the machine-generated catalogs:
python scripts/generate_skill_catalog.py
This refresh writes both the repo-owned discovery index at skill-catalog.md and the deeper governance inventory at references/skill-catalog.md.
Run the public-shape gate:
python scripts/check_private_data_policy.py
Run the public-release gate:
python scripts/check_public_release.py
cmd.exe, powershell.exe, or pwsh.exe windows may flash or remain open during startup.Default local advisor configuration:
$env:NULLCLAW_AGENT_BASE_URL="http://127.0.0.1:9000/v1"
$env:NULLCLAW_AGENT_MODEL="radeon-qwen3.5-4b"
$env:NULLCLAW_AGENT_ENABLE_LLM="1"
The advisor must remain local-only by default. The shared app-agent lane is radeon-qwen3.5-4b; Hui Hui stays reserved for avatar-specialized endpoints.
Treat http://127.0.0.1:9000/v1 as the public authoritative model plane, http://127.0.0.1:2337/v1 as the hosted large-model lane, and http://127.0.0.1:1234/v1 as a non-authoritative operator surface only.
Subagent policy:
skill-catalog.mdBOUNDARIES.mdSECURITY.mdreferences/skill-catalog.mdreferences/skill-vetting-report.mdreferences/vscode-skill-handling.mdreferences/usage-chaining-multitasking.mdreferences/OPENCLAW_NULLCLAW_THREAT_MATRIX_2026-03-11.md