← Back to Skills Marketplace
Hle Reasoning Wrapper
by
WANGJUNJIE
· GitHub ↗
· v1.0.0
644
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install hle-reasoning-wrapper
Description
Wraps HLE benchmark questions in a structured Chain-of-Thought (CoT) reasoning process. Use when answering HLE questions to ensure strict step-by-step logic...
README (SKILL.md)
HLE Reasoning Wrapper
Enforces a structured reasoning process for Humanity's Last Exam (HLE) questions.
Usage
const hle = require('./index');
const prompt = hle.formatPrompt("What is the speed of light?");
// Use prompt with LLM
const result = hle.validateOutput(llmResponse);
Logic
- Format Prompt: Injects required structure (Thought/Answer).
- Validate Output: Ensures the model followed the structure.
Usage Guidance
This skill appears to do what it claims: format HLE prompts and validate CoT-style outputs. It does create/read a cache.json file next to the skill (getCachedAnswer/cacheAnswer) — review whether you are comfortable storing past questions/answers on disk in the agent environment (this could leak sensitive prompts if the environment is shared). No network calls, external installs, or credential requests were found. If you want tighter privacy, either remove or sandbox the cache functions, or inspect/clear cache.json before/after use. Review the included code (index.js) yourself if you want to be sure no further behavior is introduced.
Capability Analysis
Type: OpenClaw Skill
Name: hle-reasoning-wrapper
Version: 1.0.0
The skill bundle is designed to format prompts for LLMs and validate their outputs, including a local caching mechanism. All file system operations are confined to a `cache.json` file within the skill's directory, which is a legitimate use case. There are no external network calls, shell executions, or attempts to access sensitive system resources. The `SKILL.md` provides benign usage instructions, and the code lacks any indicators of data exfiltration, malicious execution, persistence, or prompt injection against the OpenClaw agent itself. The direct embedding of user input into the LLM prompt by `formatPrompt` is a standard prompt engineering technique and does not constitute a vulnerability or malicious intent within the skill's own execution.
Capability Assessment
Purpose & Capability
Name/description match the provided code: functions to format prompts and validate outputs for CoT logic. Included caching functions (getCachedAnswer/cacheAnswer) are reasonable supporting features for this purpose, albeit undocumented in SKILL.md.
Instruction Scope
SKILL.md tells the agent to use formatPrompt and validateOutput; it does not mention caching. The code reads/writes a local cache.json in the skill directory, but the instructions do not instruct the agent to read arbitrary user files, call external endpoints, or access environment variables.
Install Mechanism
No install spec; this is instruction/code-only and requires only Node's builtin modules. Nothing is downloaded or written to nonstandard system locations by an installer.
Credentials
The skill requests no environment variables or credentials. The only resource access is local filesystem I/O to a cache.json file in the skill directory.
Persistence & Privilege
The module persists cached answers to cache.json within its own directory. always:false and no elevated privileges. Persistence is limited to the skill's folder, but cached content may include user questions/answers and therefore has privacy implications.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install hle-reasoning-wrapper - After installation, invoke the skill by name or use
/hle-reasoning-wrapper - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Initial release supporting structured reasoning for HLE benchmark questions.
- Provides prompt formatting to enforce Chain-of-Thought (Thought/Answer) structure.
- Validates model outputs for compliance with the required reasoning steps.
- Simple API: `formatPrompt` to structure prompts and `validateOutput` to check responses.
Metadata
Frequently Asked Questions
What is Hle Reasoning Wrapper?
Wraps HLE benchmark questions in a structured Chain-of-Thought (CoT) reasoning process. Use when answering HLE questions to ensure strict step-by-step logic... It is an AI Agent Skill for Claude Code / OpenClaw, with 644 downloads so far.
How do I install Hle Reasoning Wrapper?
Run "/install hle-reasoning-wrapper" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Hle Reasoning Wrapper free?
Yes, Hle Reasoning Wrapper is completely free (open-source). You can download, install and use it at no cost.
Which platforms does Hle Reasoning Wrapper support?
Hle Reasoning Wrapper is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Hle Reasoning Wrapper?
It is built and maintained by WANGJUNJIE (@wanng-ide); the current version is v1.0.0.
More Skills