Manusilized

Enhances OpenClaw agents with real-time streaming, Markdown tool-call recovery, and extended support for latest reasoning models.

Security Scan

The skill's files, instructions, and install helper are consistent with its stated purpose (patching OpenClaw's Ollama provider for streaming, markdown tool-call fallback, and model heuristics), but it is an invasive core patch from an unknown source so proceed cautiously.

Purpose & Capability

The name/description promise real-time streaming, Markdown tool-call recovery, and model heuristics; the package contains TypeScript patches (ollama-stream.ts, ollama-models.ts), a README, and an install script that copies those files into OpenClaw's src/agents. There are no unrelated env vars or binaries requested. This is coherent with the stated purpose.

Instruction Scope

SKILL.md instructs the user to either wait for an upstream PR or manually replace two core files and rebuild. The included install-patch.sh performs local file backups and copies the patch files; there are no instructions to read unrelated system files, exfiltrate secrets, or call unexpected remote endpoints from the installer. The runtime instructions are limited to modifying OpenClaw core code, which is exactly what the skill claims to do.

Install Mechanism

There is no remote download/install spec — the skill ships patch files and an install script that copies them into the user's OpenClaw tree. This lowers supply-chain risk compared to fetching arbitrary code, but it is still an invasive change because it replaces core agent files. The install script does create .bak backups, which is good practice.

Credentials

The skill does not request environment variables, credentials, or config paths. The code uses local API base URLs (Ollama native endpoints) and normal fetch calls; nothing in the files demands unrelated secrets or external credentials. This is proportionate to its function.

Persistence & Privilege

The skill does not set always:true and is user-invocable (defaults). However, applying the patch permanently modifies the agent runtime by replacing core files. That is inherently higher-privilege than a non-invasive skill, but it's an explicit, user-initiated change rather than an autonomously persistent modification.

Assessment

This package is internally consistent with its stated goal of patching OpenClaw's Ollama provider, but it modifies core agent code — treat it like any other high-impact patch. Before installing: (1) review the full patch files line-by-line (you saw a truncated file in the registry listing — inspect the rest locally), (2) verify the upstream repository and author (the registry metadata lacks a homepage and the source is 'unknown'), (3) test the patch in a disposable clone or non-production environment and keep the .bak files the installer creates, (4) prefer the official upstream PR merge over local patching, and (5) search the patched code for unexpected network endpoints or telemetry if you need high assurance. If you aren't comfortable auditing the TypeScript patches yourself, decline or wait for the official PR to be merged and reviewed.

Manusilized: The OpenClaw Supercharger

Manusilized is not just a skill; it's a comprehensive upgrade for your OpenClaw agent, bringing enterprise-grade reliability and a "Manus-like" silky-smooth experience to open-source models (like Qwen, GLM, DeepSeek).

🌟 Key Features

1. Incremental Streaming (Silky Smooth Experience)

   • No more staring at a blank screen while reasoning models think! Manusilized introduces real-time text_delta streaming for Ollama models.
   • You get a live typewriter effect, perfectly mirroring the behavior of closed-source giants.

2. Markdown Tool-Call Fallback (The "Amnesia" Cure)

   • Open-source models often struggle with native JSON Function Calling, outputting tool calls embedded within Markdown code blocks.
   • Manusilized includes a fault-tolerant adapter that automatically detects, extracts, and corrects these embedded tool calls, drastically improving success rates.

3. Extended Reasoning Model Support (2026 Ready)

   • Natively recognizes and optimizes for the latest generation of reasoning models: qwen3, qwq, glm-5, kimi-k2.5, deepseek-v3, marco-o1, and skywork-o.

🛠️ Installation & Setup

Since Manusilized modifies the core agents architecture of OpenClaw, it cannot be installed merely as a standard skill folder.

Option 1: The Official PR (Recommended)

We have submitted these upgrades as a core PR to the OpenClaw repository. 👉 View and Upvote the PR on GitHub

Option 2: Manual Patch

If you want to experience Manusilized immediately before the PR is merged, you can apply our patch directly to your OpenClaw installation:

1. Clone the OpenClaw repository.
2. Download the Manusilized patch files from our GitHub repository.
3. Replace the corresponding files in src/agents/ (ollama-stream.ts and ollama-models.ts).
4. Rebuild OpenClaw.

💡 Why Manusilized?

Our mission is to democratize the "perfect agent" experience. By bridging the gap between open-source models and enterprise frameworks, Manusilized ensures that anyone can run a top-tier AI agent locally or in the cloud without relying on expensive API keys.

---

Created by Manus AI.