GRREAT / Agent skill

Keep the plan.
Carry the work forward.

Give your agent a method for planning, doing, and picking up where you left off. Your project’s memory lives in linked Markdown you can read and edit.

Install the skill Already working on something?

Local use needs no account. App sync is optional.

Two halves. One continuing project.

First half / Plan mode

Find the direction.

Goals: What matters, what is in scope, and what success looks like.
Research: What is known, the evidence behind it, and the questions still open.
Roadmap: Milestones, dependencies, and how each result will be checked.

Second half / Execution

Make progress visible.

Execution: The current action, completed work, blockers, and verification evidence.
Analysis: What the results mean for the goal and which decisions need to change.
Time: Estimates and observed effort, with unknowns kept explicit.

The handoff: Plan mode develops the first half in conversation. Once execution is authorized, the agent saves the plan and maintains the vault as it works. Evidence feeds back into Goals, Research, and Roadmap whenever the work changes the picture.

01 / Install once

Available across your projects.

Install grreat from the public skill repository, or use the versioned download below. Both contain the same 1.2.0 skill.

Download GRREAT 1.2.0 and its SHA-256 checksum. Verify the archive before extracting it. It contains a folder named grreat with its instructions and helpers.
Ask your agent to install that folder at user scope, or place it at ~/.agents/skills/grreat/ for Codex. The file SKILL.md should sit directly inside it.
Open a project and invoke $grreat. If it does not appear, restart Codex. For another compatible agent, use its skill installation and invocation instructions.

The bundled helpers need Node.js 22 or later. Local tracking needs no npm install or account.

For an update, check metadata.version in your installed SKILL.md, back up that skill folder, and preserve personal changes before replacing it. Keep one active copy. Project notes and sync checkpoints stay in your project. See OpenAI’s skill documentation for Codex’s user scope and discovery behavior.

Verify the download

Save the archive and checksum in the same folder, then run this there on macOS or Linux:

shasum -a 256 -c grreat-1.2.0.tar.gz.sha256

Continue when the archive reports OK. On Windows, use Get-FileHash -Algorithm SHA256 on the archive and compare it with the downloaded checksum.

02 / Start where you are

A new goal or work already underway.

One prompt to get started

Open the project in your agent and paste this. For a new project, replace “adopt this ongoing project” with your goal.

Install and adopt this project

Install or update the GRREAT skill for my user scope from https://grreat.ca/skill. Verify the release checksum, preserve my existing skill customizations, and keep one active installation.

Use $grreat to adopt this ongoing project. Inspect its existing plans and work before changing anything. Reuse its vault, or create docs/grreat. Help me identify the first meaningful Goal, success signals, and next action. Keep Goals, Research, and Roadmap separate from Execution, Analysis, and Time.

Keep this local. App sync is optional; offer it after adoption and send notes only if I request it. If I choose sync, use the installed skill’s secure credential flow, preview all canonical project records, confirm the authenticated destination, sync those records, and report the first Goal’s receipt and Continue link.

Planning and execution stay local until you choose to connect the app.

Adopt an ongoing project

Use $grreat to adopt this ongoing project.

The agent reads the project and its existing plans, records what is known, and identifies the next action. It preserves existing records and separates historical claims from newly verified results.

Plan something new

Use $grreat to plan this project: [your goal].

Switch to Plan mode first. Develop the outcome, research questions, and milestones. When you are ready, switch to execution and ask the agent to implement the plan with GRREAT tracking.

Resume the work

Use $grreat to resume this project.

The agent reads the current state and relevant linked records, checks them against the project, then continues from the next action. You can also ask it to review progress against the goal.

03 / Keep the memory

Notes that connect.

The default home is docs/grreat/ inside your project. Adoption reuses an existing GRREAT vault when one is present. A compact current-state note points to the goal, milestone, next action, blockers, and recent evidence.

Goals, research, decisions, and work journals connect through [[wiki links]]. Open the folder as an Obsidian vault to follow links and backlinks, or use your usual Markdown editor. The pages and journals follow a Logseq-style organization; this is a Markdown vault, not a Logseq database import.

The agent updates the records after meaningful progress and leaves enough context for a fresh session to resume. Checks stay attached to the state they actually verified.

04 / Optional connections

Add what the project needs.

Connect to the GRREAT app

The bundled sync-cli.mjs can sync your Markdown records to your workspace. Ask the agent to set up app sync using the skill’s instructions. Your local vault remains the source of truth; sync uses authenticated access.

Agent connections and workspace access

Search a larger project

If you already use zvec-grep, it can help discover related material by meaning. It is optional. The agent still reads the underlying files and follows explicit links before treating a result as evidence.