Vibeleaderboard
← Back to Tools
Index / agent

Linear Coding Agent Harness

https://github.com/coleam00/Linear-Coding-Agent-Harness
Visit github.com
Category
AI Agents
Rank
No. 1123Tools index
Pricing
Open Source
Type
AGENT
Builder
coleam00
GitHub
229 stars
Date

About

Harness that turns a Linear issue into work for an autonomous coding agent — assign, code, PR, done.

What it does

This Python harness runs fresh Claude coding sessions against a shared Linear project. An initializer converts an application specification into detailed issues and project scaffolding. Later sessions choose priority work, verify earlier features through Puppeteer, implement changes, and preserve context in issue comments, statuses, and a dedicated handoff issue.

Why it's ranked here

The design tackles a real weakness of long-running agents: context disappears between sessions. Linear becomes durable memory and a visible control surface, while browser checks and git commits add discipline. The result is compelling as a concrete reference implementation, though its fixed assumptions and narrow toolchain make it less suitable as a general automation framework.

What's good

Every session begins with explicit orientation and checks existing work before starting another feature. Issues carry acceptance steps, comments record implementation details, and interrupted work remains visible through status. The security setup combines sandboxing, project-scoped file permissions, an allowed-command list, and extra validation for sensitive development commands.

Tradeoffs

Setup requires Claude Code, its Python SDK, a Claude OAuth token, a Linear workspace, and a read-write Linear API key. Initialization creates 50 issues and may take over ten minutes. The loop retries errors indefinitely unless iterations are capped. Browser verification targets web applications, while command permissions are tuned narrowly for Node development.

How to use it well

Use it as a starting point for teams already managing substantial web builds in Linear and willing to encode detailed acceptance tests. Customize the application specification, issue count, model, and command policy before running. It earns a place in long, restartable implementation work. It does not replace product planning, general CI, or support for non-browser verification.

Technical notes+

autonomous_agent_demo.py provides the argparse CLI, checks both required tokens, and places relative targets under generations/. agent.py recreates ClaudeSDKClient context each iteration and selects initializer or continuation prompts using the marker state read by progress.py. client.py configures Linear over streamable HTTP, Puppeteer over stdio, project-scoped permissions, sandboxing, a Bash pre-tool hook, and max_turns=1000. security.py parses commands with shlex plus regular-expression segmentation, checks ALLOWED_COMMANDS, and applies dedicated validation to pkill, chmod, and init.sh. test_security.py exercises command extraction, allowed commands, blocked commands, and sensitive-command validators. requirements.txt declares only claude-code-sdk>=0.0.25.

Observed

License
MIT License
Primary language
Python
Install surface
Global Claude Code CLI installation plus pip installation from requirements.txt
Interface
Command-line application with project directory, iteration limit, and model options
External interfaces
Linear MCP over streamable HTTP and Puppeteer MCP over stdio
Authentication
Requires a Claude Code OAuth token and a Linear API key
Testing structure
Repository includes a standalone Python security test script

Read from README.md, requirements.txt, agent.py, client.py, prompts.py, progress.py, security.py, linear_config.py, test_security.py, autonomous_agent_demo.py, LICENSE, prompts/app_spec.txt, prompts/coding_prompt.md, prompts/initializer_prompt.md.

What it can do

  • Convert Linear issue to coding task

    Linear issue → Structured coding task for autonomous agent

  • Automatically assign coding tasks to agent

    Linear issue or coding task → Task assignment to coding agent

  • Generate code from issue requirements

    Linear issue with requirements → Source code implementation

  • Create pull request from generated code

    Generated code and original issue → Pull request in code repository

  • Automate end-to-end development workflow

    Linear issue → Completed pull request ready for review

Tags

linearcoding-agentautonomouspython

Tech Stack

Python

Comments (0)

No comments yet

Editorially curated, with community endorsements as a secondary signal. Corrections welcome.