Smart Router Omni

Universal smart routing skill that chooses the best installed skill or skill chain across mixed environments, and automatically applies OpenClaw-aware routin...

Security Scan

Instruction-only router that is internally consistent with its stated purpose: it inspects skill metadata and workspace markers to pick or chain skills and does not request external installs, binaries, or credentials.

Purpose & Capability

Name and description match the runtime instructions. The skill's declared goal (choose best skill or chain, support OpenClaw-aware mode) explains why it would read SKILL.md frontmatter, agents/openai.yaml for shortlisted candidates, and OpenClaw workspace markers.

Instruction Scope

Runtime instructions ask the agent to read SKILL.md frontmatter, agents/openai.yaml (for shortlisted candidates), and OpenClaw workspace indicators (e.g., MEMORY.md, .openclaw/, memory/, bank/). This is appropriate for a router, but those environment files can contain sensitive workspace-specific data (e.g., memory contents or config). The skill does not instruct arbitrary system file reads or network exfiltration, and it recommends conservative shortlisting.

Install Mechanism

No install spec and no code files beyond documentation; nothing will be written to disk or downloaded during install. This is the lowest-risk install model.

Credentials

The skill requests no environment variables, credentials, or external endpoints. It does rely on reading workspace files and skill metadata; because workspace memory/config may be sensitive, ensure you accept the agent's ability to inspect those local artifacts before enabling autonomous routing.

Persistence & Privilege

always is false and there is no install-time persistence. The agent is allowed autonomous invocation by default (platform normal). The skill does not request elevated privileges or modify other skills' configurations.

Assessment

This skill is an instruction-only router and appears coherent for its stated purpose: it will read SKILL.md frontmatter, agents/openai.yaml for shortlisted candidates, and OpenClaw-style workspace markers to choose or chain skills. There are no downloads, no required credentials, and no system-wide persistence. Before installing or enabling autonomous invocation, consider whether you are comfortable with the agent having read access to workspace files that may include memory or configuration (MEMORY.md, .openclaw/, bank/, etc.). If that is a concern, run the skill in a restricted workspace or deny autonomous invocation until you review its behavior manually. Note: the skill's proprietary license forbids copying or reverse-engineering of this bundle, but that is a legal constraint separate from runtime behavior.

Start broad, then specialize. Default to universal routing, and switch into OpenClaw-aware mode when the environment supports it.

Load the right reference

• Read references/scoring-rubric.md before scoring candidates.
• Read references/ambiguity-and-fallbacks.md when confidence is weak or constraints are missing.
• Read references/environment-detection.md before deciding whether to stay in universal mode or enter OpenClaw mode.
• Read references/chain-patterns.md when the request obviously spans several phases.
• Read references/research-notes-2026-03.md when you need the design rationale.

Workflow

1. Normalize the request into a task card:
   • goal
   • expected artifact
   • domain
   • required actions
   • constraints
   • environment dependencies
2. Detect the environment using references/environment-detection.md:
   • universal mode
   • OpenClaw-aware mode
3. Scan visible skill roots conservatively:
   • read SKILL.md frontmatter first
   • inspect agents/openai.yaml only for shortlisted candidates
4. Apply hard filters before ranking:
   • missing tools
   • missing login or auth
   • wrong output type
   • platform mismatch
   • safety or policy mismatch
5. Score viable candidates with references/scoring-rubric.md.
6. If one skill can finish the task, recommend that skill.
7. If the task spans phases, recommend a short chain using references/chain-patterns.md.
8. If confidence is low, ask compact clarification questions instead of forcing a route.
9. Output:
   • recommended skill or chain
   • confidence
   • why it won
   • prerequisites
   • fallbacks
   • clarifying questions if needed

Routing policy

• Prefer explicit capability claims in description over name similarity.
• Prefer specialist skills when the task, artifact, and dependencies are clear.
• Prefer workflow skills when the request is end-to-end.
• Prefer the smallest route that can finish the task well.
• In OpenClaw-aware mode, treat browser, publishing, memory, and account-bound skills as dependency-sensitive.
• Abstain and clarify when the top candidates are close.

Standard output

[Routing Decision]
Mode: universal | openclaw-aware
Request: ...
Recommended skill: ...
Confidence: high | medium | low
Why it fits: ...
Missing checks or prerequisites: ...
Fallbacks: ...
Suggested chain: ...
Clarifying question(s): ...

Guardrails

• Do not rely on a fixed handwritten route table as the primary method.
• Do not read every full SKILL.md before shortlisting.
• Do not recommend unavailable skills.
• Do not hide uncertainty when the top candidates are close.
• Do not force one skill when a chain is clearly better.

Exit condition

Finish with one recommended skill or skill chain, an explicit mode, a confidence level, key reasons, fallbacks, and any blocking unknowns.

---

Copyright & License

Copyright (c) 2026 龙虾 (Lobster)

All Rights Reserved.

This skill is proprietary and confidential. The source code, algorithms, documentation, and any accompanying materials are protected by copyright law and international treaties.

You may NOT:

• Copy, modify, or distribute the source code or documentation
• Create derivative works based on this skill
• Use this skill as a template or baseline for developing competing products
• Reverse engineer, decompile, or disassemble any components
• Remove or alter any proprietary notices or copyright labels

You MAY:

• Install and use this skill in your OpenClaw installation
• Receive updates and support as provided by the author

For licensing inquiries or custom development, contact the author directly.