.agentsGitHub

Make AI coding work survive the next thread.

dot-agents stores work items, plans, and verification in your repo’s plain .agents/ folder, so the next thread can continue without reconstructing the work.

Install with one command

Add it to your repo.

Run this from your repository root. It creates AGENTS.md and .agents/, and preserves your work when you sync.

terminal — repo rootbash
 curl -fsSL https://raw.githubusercontent.com/colmarius/dot-agents/main/install.sh | bash

Reviewcurl -fsSL https://raw.githubusercontent.com/colmarius/dot-agents/main/install.sh | less

Pin v0.4.0curl -fsSL https://raw.githubusercontent.com/colmarius/dot-agents/main/install.sh | bash -s -- --ref v0.4.0

Then tell your agent: Run adapt.

The working loop

A small protocol for long-running work.

Keep only the context another thread needs to continue.

  1. 01

    Frame

    Open a work item

    Capture the goal, current status, and next action in index.md.

  2. 02

    Context

    Add what earns its place

    Add research, requirements, or a plan only when the work needs it.

  3. 03

    Execute

    Work from the repo

    Implement from durable repo context, in the current thread by default.

  4. 04

    Prove

    Record the evidence

    Record what you verified and the exact next action.

What lands in your repo

Plain files. One place to resume.

A new thread reads index.md, then loads only the plan, research, or progress it needs.

Why durable context?
your-project/tracked
├── AGENTS.md       # project instructions
└── .agents/
    ├── work/       # durable work items
    ├── research/   # reusable findings
    ├── references/ # local reference repos
    ├── skills/     # repeatable workflows
    └── scripts/    # safe sync updates
work/feature/user-authentication/index.md● saved
# User authentication

Status: planned
Category: feature
Updated: 2026-06-22

## Why
Users need secure access across sessions.

## Summary
Add auth flows and session persistence.

## Next Action
Implement Task 1 from plan.md.

Start with these prompts

Ask for what you need.

Use these prompts as written. Add detail only when it helps.

Run adapt

Fill AGENTS.md with the repo’s commands and conventions.

Create a new work item for ...

Start durable context under .agents/work/.

Create a plan for ...

Turn stable context into implementation-ready tasks.

Write a handoff prompt for ...

Prepare a scoped prompt when another thread would help.