anolisa/src/agentsight/dsh-plugin

One-Line Positioning

This is a bridge plugin that connects AgentSight's observability capabilities to DSH (DeepSeek Harness). It only registers a single tool called agentsight_status, allowing running Agents to query whether the plugin is activated and its current version—essentially exposing AgentSight's "self-introduction" entry point within DSH.

Core Capabilities

• Injects a Cordis plugin named agentsight into DSH, depending on the host's tools extension
• Registers the built-in tool agentsight_status for Agents to call during sessions
• When called, returns the plugin name, version number (dynamically read from package.json), activation status, and current capability list
• Outputs a startup log via ctx.logger.info when the plugin finishes loading, marking the registered version number and tool name

Technical Implementation

• Language: TypeScript (ESM, target ES2022)
• Key Dependencies: @deepseek-ai/cordis ^4.0.1 (plugin host framework), @deepseek-ai/dsh-tools ^0.1.0-rc.2 (DSH tool registration extension)
• Architecture Pattern: Standard Cordis plugin pattern — declares insertion points via insert in cordis.patch.yml, executes apply(ctx) at runtime to register tools to ctx.tools
• Entry File: src/index.ts (49 lines, all logic)

Use Cases

Install this plugin when users have already deployed the AgentSight main program on their machines (responsible for eBPF tracing, token statistics, Dashboard, etc.) and want Agents within DSH to be aware of AgentSight's presence. Typical usage is for Agents to call agentsight_status for self-check at the start of each session or when asked "what observability tools do I have installed," rather than hardcoding version numbers in conversations.

Prerequisites and Compatibility

Installation

dsh plugin --profile web add github:alibaba/anolisa/src/agentsight/dsh-plugin

Configuration

No additional configuration required for this plugin.

FAQ

Q: Can this plugin collect Agent's LLM calls, token consumption, or network traffic?

A: No. This plugin only registers an agentsight_status tool, does not start any probes, does not read network data, and does not write to databases. The actual collection work is done by the AgentSight main program, which needs to be deployed separately.

Q: Do I need additional configuration after installation?

A: No. The plugin only injects and registers a built-in tool via Cordis. It has no Schema configuration items, does not read environment variables, and does not read/write external files. Install and use immediately.

Q: What is its relationship with the AgentSight main program?

A: It is a lightweight bridge layer between DSH and AgentSight, only exposing status query capabilities; the actual eBPF data collection, SQLite storage, HTTP API, and Dashboard are provided by the AgentSight main program, which needs to be deployed separately.

Q: Will uninstalling affect already collected data?

A: The plugin itself does not persist any data, so uninstalling it is safe; it does not touch the AgentSight main program's SQLite, configuration, or trace records.

Q: Can it run on macOS or Windows?

A: The plugin code itself is cross-platform TypeScript/Node.js code, but the AgentSight main program it bridges depends on Linux kernel's eBPF capabilities. If only this plugin is enabled without the main program installed, the tool will always report active=true but capabilities will only contain status.

Q: Where does the version returned by calling agentsight_status come from?

A: It is read via fs.readFileSync reading its own package.json's version field when the plugin apply() executes, so the returned value is always the version bundled into the plugin at build time.

Getting Started

Beginner — only one tool, no configuration, no side effects, install and use.

Known Issues and Limitations

• The capabilities field of the agentsight_status tool is currently hardcoded to return ['status'] and does not dynamically change based on whether the host has the AgentSight main program installed, so the return value cannot be used to determine if the main program is running
• The plugin does not provide interfaces for querying the AgentSight main program's version, running status, or collection channel health; callers needing more detailed information need to query the main program's own HTTP API instead
• Since apply() depends on the host exposing the ctx.tools extension, loading will fail in host environments that do not meet inject: ['tools']

---

ANOLISA is a server-side operating layer for AI Agent workloads. It addresses three practical constraints of Agent execution: terminal entry, Token cost, and execution environments. Keep the Shell, Agent framework, and sandbox you already use. ANOLISA CLI provides a single installation entry point, while each capability can be enabled independently.

New to ANOLISA? Choose your first outcome in the Quick Start →

Components

What it solves

01 · AGENT INTERFACE

Let the Agent work directly in the terminal

cosh-ng is an AI-native Linux terminal: it keeps familiar Bash/Zsh behavior, then adds an Agent that can understand intent, use tools and Skills, and ask for approval before risky work. Shell commands and natural language share one terminal instead of forcing users into a separate chat application.

Get started with cosh-ng →

02 · CONTEXT EFFICIENCY

See where Tokens go and cut waste before it reaches the model

Token-less removes redundancy from tool schemas and responses before they reach the model. Agent Memory reuses useful context across sessions. SkillFS keeps the current Skill view focused and makes other Skills discoverable when needed. AgentSight shows where Tokens are spent.

See an Agent run from the kernel up

On Linux, AgentSight uses eBPF to observe an Agent without changing its code. Follow user input through model and tool calls, with Token use and sub-agent branches in the same view.

Open the AgentSight guide →

Try Token-less with Claude Code in 3 minutes

Install Token-less and connect it to Claude Code:

curl -fsSL https://get.agentic-os.sh | bash
export PATH="$HOME/.local/bin:$PATH"
anolisa install tokenless
anolisa adapter enable tokenless claude-code

Restart Claude Code, run one tool-heavy task, then inspect the result:

tokenless stats summary
tokenless stats list --limit 5

Open the full Token-less Quick Start → · Read the user manual

_{In one observed coding task, Token-less saved 317K Tokens (40.5%), based on AgentSight measurements. Results vary by workload.}

debug and trace are dropped by the field blacklist, metadata as null, and tags / extra as empty values. Compression runs between the Agent and the model, so no Agent framework code changes. Dropped array items stay retrievable through a <<tokenless:KEY>> marker, which keeps the compression reversible.

Savings apply to the tool responses entering the context, not to the whole session bill. The Token-less user manual explains how to estimate the effect for a given workload.

03 · EXECUTION RUNTIME

Give every Agent execution a boundary and a way back

ANOLISA is building out the Agent execution environment: Agent Sec Core isolates risky operations, and ws-ckpt keeps recovery points for workspace changes.

Catch a changed Skill before it runs

When a signed Skill changes, the Agent reports drifted before using it again. A rescan records blocking findings as deny.

Try the Agent demo → · Skill Ledger guide

Choose a runtime or security starting point → · Start with ANOLISA CLI

Install

ANOLISA CLI is the common installation entry point. cosh-ng is installed in system mode; Token-less and other capabilities can be added independently.

curl -fsSL https://get.agentic-os.sh | bash

sudo anolisa --install-mode system install cosh-ng
anolisa install tokenless

Run cosh to enter the AI-native terminal. Token-less can also optimize tool calls from an existing Agent without changing its framework.

Read the Quick Start →

Documentation

Quick Start · Installation · User Guide · Troubleshooting · Build from Source · Changelog

Community

• Open an issue for bugs and feature requests.
• Read CONTRIBUTING.md before submitting a pull request.
• Report vulnerabilities through the Security Policy.

License

ANOLISA is released under the Apache License 2.0.