Analysis updated 2026-05-18
Set a long-running objective so an AI coding agent keeps working across many turns without re-prompting.
Cap how many turns an agent can use on a task with an automatic wrap-up when the budget runs out.
Add auto-continuation hooks to Claude Code so refactors and test-writing tasks run to completion.
| secemp9/goal | jerlinn/jerlin-weread | jonluca/dotfiles | |
|---|---|---|---|
| Stars | 6 | 6 | 6 |
| Language | Shell | Shell | Shell |
| Last pushed | — | — | 2024-06-02 |
| Maintenance | — | — | Dormant |
| Setup difficulty | moderate | — | easy |
| Complexity | 2/5 | — | 1/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Full experience requires uv, jq, and bash 4+, plus running install.sh to register hooks.
Goal is a tool that gives AI coding assistants, such as Claude Code, Cursor, and OpenCode, a persistent memory of what they are trying to accomplish across multiple exchanges. Without it, an AI agent stops after one response and forgets the overarching objective, so you have to manually re-prompt it to keep going. Goal solves this by letting you declare an objective once, and the agent works toward it across many turns automatically. The tool installs as an MCP server, MCP being a standard way for AI tools to communicate with assistants. Once installed, it exposes three core actions: creating a goal, checking current status, and marking a goal complete. Goal state is stored as a JSON file per project, tracking the objective text, current status (active, paused, budget limited, or complete), turns used, and elapsed time. For Claude Code specifically, Goal also installs hooks, small scripts that fire at key moments in a session. One injects a goal reminder at the start of each turn. Another monitors turn usage and warns when approaching the budget limit. A third is the auto-continuation engine: after each turn ends, it checks whether the goal is still active and automatically starts another turn without you re-prompting. Other editors get the MCP tools but not auto-continuation, so they require manual prompting to continue. You would use this when you have a large, multi-step task, such as refactoring a module, writing a test suite, or debugging a complex issue, and want an AI assistant to keep working without constant hand-holding. Written in Shell, ported from the goal architecture in Codex CLI, and released under the MIT license.
An MCP tool that gives Claude Code, Cursor, and OpenCode a persistent goal that auto-continues across turns until finished.
Mainly Shell. The stack also includes Shell, Python, MCP.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice (MIT license).
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.