UltraGoal makes your agent transform a lazy goal into an Ultra Goal.
Goal mode is amazing, but half the battle is writing your goal with sufficient detail to be effective. The agent turns your rough intention into a real working contract: expand on what "done" means, what's in scope, how to verify it, how much autonomy it has, and when to stop. And it does this using the context you already have available to your agent, so there's less guessing and more doing.
A normal goal asks the agent to start. An Ultra Goal makes the agent define the target before it starts moving.
You invoke it as a slash command:
/ultragoal fix the settings bug
The slash command is the replacement for a normal goal prompt. Everything after
/ultragoal is raw intent. UltraGoal makes the agent sharpen it before doing the
work.
lazy goal -> Ultra Goal -> agent executes
Want Codex, Claude, or another agent to install it for you? Paste this into a fresh thread:
Set up UltraGoal from https://github.com/justinwetch/UltraGoal.git. Once it is installed, tell me how to invoke it in this environment.
This repo is intentionally small.
UltraGoal/
├── README.md
├── banner.png
├── LICENSE
├── ultra-goal-skill.zip
└── .agents/skills/ultra-goal/
├── SKILL.md
├── agents/openai.yaml
└── references/goal-rubric.md
The zip is the portable artifact. The .agents/skills/ultra-goal/ folder is the
same skill unpacked, kept in the repo so agents and humans can inspect it without
unzipping anything. It also keeps changes reviewable in Git.
There are no product notes, render pipelines, videos, frame dumps, or development scratch files here. Those belong somewhere else.
Use UltraGoal on a rough goal:
/ultragoal fix the bug where settings don't save
The agent should:
- Treat everything after
/ultragoalas raw intent. - Rewrite it into a stronger persistent goal.
- Set that upgraded goal as the active goal when the runtime exposes goal tools, or proceed from it as the working contract when it does not.
- Work from the upgraded goal without asking a preflight question.
An upgraded goal carries the parts people usually forget:
- objective
- scope
- success criteria
- verification
- wall-clock constraints
- autonomy boundaries
- completion rule
User time instructions are part of done. When the user gives a deadline, time limit, minimum active-work duration, or scheduled work window, UltraGoal carries that into the upgraded goal. Minimum-duration instructions require active work, iteration, verification, documentation, or blocker analysis for the prescribed time; idle waiting, sleeping, artificial delay, or burning time to satisfy the clock is a failure.
UltraGoal is deliberately narrow. It is the handoff between human intent and agent goal state. It is not a prompt coach, an interview flow, a planning preview, or an approval step.
The zipped skill is here:
ultra-goal-skill.zip
The unzipped skill is here:
.agents/skills/ultra-goal/
Both contain the same skill.
To regenerate the zip after editing the source:
cd .agents/skills
zip -qr ../../ultra-goal-skill.zip ultra-goalMIT.
