← Back to Skills Marketplace
Surfagent Perception
by
AgentOSsoftware
· GitHub ↗
· v1.0.0
· MIT-0
107
Downloads
0
Stars
1
Active Installs
1
Versions
Install in OpenClaw
/install surfagent-perception
Description
Agent vision for web pages — scene summaries, attention-ranked elements, annotated screenshots, and state diffing via SurfAgent's perception engine.
Usage Guidance
This skill is instruction-only and expects a local SurfAgent daemon (localhost:7201) and a managed Chrome (9222). Before installing: (1) verify you trust the SurfAgent daemon you run locally — the skill will capture full page DOM and screenshots which can include sensitive data; (2) confirm why 'node' is declared as a required binary if you don't have a node-based SurfAgent client — it may be a packaging artifact; (3) run SurfAgent in an isolated environment if you plan to visit sensitive sites, and avoid giving the agent access to pages with credentials or personal data unless you understand redaction behavior; (4) if you need higher assurance, ask the publisher for source code or a specification of the protocol (endpoints, authentication, and data retention) and confirm the homepage/service is legitimate.
Capability Analysis
Type: OpenClaw Skill
Name: surfagent-perception
Version: 1.0.0
The surfagent-perception skill bundle contains metadata and documentation for a web-page analysis toolset. The SKILL.md file provides functional instructions for an AI agent to interact with a local SurfAgent daemon for tasks like scene summarization and element ranking. No executable code was provided for analysis, and the documentation contains no malicious prompt injections, exfiltration instructions, or suspicious obfuscation.
Capability Assessment
Purpose & Capability
The SKILL.md consistently describes a perception engine that runs as a local SurfAgent daemon (port 7201) backed by a managed Chrome instance (port 9222). The described tools (surf_perceive, surf_annotate, surf_scene_diff) and their outputs (scene summaries, annotated screenshots, state tokens) align with the skill's name and description.
Instruction Scope
Instructions are limited to calling a local SurfAgent service to capture DOM, annotated screenshots, and diffs. They do not ask to read arbitrary files or unrelated environment variables. Important operational detail: the agent is expected to connect to localhost:7201 and interact with Chrome (CDP), which can capture page content including any sensitive information displayed on pages (passwords, auth tokens, personal data). The SKILL.md does not describe any redaction or privacy safeguards.
Install Mechanism
No install spec and no code files — this is instruction-only, so nothing gets written to disk by the skill itself. That is the lowest install risk.
Credentials
The skill declares no required environment variables or credentials, which is proportionate. One minor inconsistency: SKILL.md metadata lists 'node' as a required binary even though the instructions talk only to a local SurfAgent daemon (HTTP + Chrome). Requiring 'node' may be unnecessary or a packaging artifact; it does not appear to grant extra access, but it is an unexplained dependency.
Persistence & Privilege
The skill does not request always:true and is user-invocable. Autonomous invocation (default) is allowed, which is expected for skills. Note: autonomous use combined with the ability to control a local browser has a substantial operational privilege (it can observe and interact with pages), but that is coherent with the skill's purpose.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install surfagent-perception - After installation, invoke the skill by name or use
/surfagent-perception - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Initial release of SurfAgent Perception providing advanced web page understanding and visual annotation tools.
- Introduces `surf_perceive` for scene summaries, attention-ranked elements, spatial clusters, and annotated screenshots.
- Adds `surf_annotate` for quick, legend-backed visual references of interactive elements.
- Implements `surf_scene_diff` for efficient state change detection after user actions.
- Core loop enables perceive→act→diff→repeat workflow for robust agent-page interaction.
- Features element attention ranking, visual clustering, delta reporting, and color-coded annotation support.
Metadata
Frequently Asked Questions
What is Surfagent Perception?
Agent vision for web pages — scene summaries, attention-ranked elements, annotated screenshots, and state diffing via SurfAgent's perception engine. It is an AI Agent Skill for Claude Code / OpenClaw, with 107 downloads so far.
How do I install Surfagent Perception?
Run "/install surfagent-perception" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Surfagent Perception free?
Yes, Surfagent Perception is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Surfagent Perception support?
Surfagent Perception is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Surfagent Perception?
It is built and maintained by AgentOSsoftware (@agentossoftware); the current version is v1.0.0.
More Skills