leader - Turn a One-Sentence Idea into a Task Brief an Agent Can Complete Independently

Skill Overview

leader is a Claude Code skill that first researches your one-sentence idea, then breaks it down into a goal-oriented task brief that an AI agent can execute independently. You can paste it directly into the /goal command to run.

Use Cases

1. You have a project idea and want an agent to complete it independently, but don’t know how to write a reliable goal prompt.
2. The idea is too vague, so you need to test the codebase first, verify the relevant information, and then break it down into an executable, verifiable goal.
3. The task is too large to finish in one go, and you want to split the project into several task briefs for multiple agents to work on in parallel.

Core Features

1. Research first, then write the brief: Test the codebase hands-on (whether commands actually exist, what the baseline numbers are, and how far the documentation diverges from reality), and consult industry resources online when necessary. Anything that cannot be verified is explicitly marked as “Assumption, unverified.” Every command included in the brief is personally tested to prevent the agent from receiving hallucinated commands.
2. One round of questions to fill in key decisions: Ask only about issues that cannot be determined through research and that would change the task brief, such as trade-offs, acceptance criteria, risk tolerance, and time limits. Each question includes 2–4 options and a recommendation. If you do not make a decision, the default is used and clearly marked as “Decision made on your behalf”—never decided silently.
3. Cheat-proof acceptance criteria: The task brief includes hard completion requirements: baseline metrics must not regress; shortcuts such as skipping tests are explicitly prohibited; evaluation criteria are frozen; and changes must be rolled back if results worsen. It also includes progress tracking and checkpoint/resume procedures so another agent can take over without repeating completed work. After execution, the author personally reruns the acceptance checks and gives you a plain-language report in no more than five lines.
4. Parallel decomposition across multiple agents: With your approval, large tasks are split into multiple task briefs with non-overlapping scopes and clearly assigned integration points, preventing agents from stepping on each other’s work. Each brief includes the same global instructions describing the division of labor and handoff points.

Frequently Asked Questions

What is the leader skill?

leader is a skill for people who come up with ideas: you provide the idea and make the decisions, while it handles the research, questions, task-brief writing, and acceptance checks. The output is a task brief of no more than 4,000 characters. The execution agent can complete it independently without your supervision or any commands from you during the process.

How do I use a task brief generated by leader?

In the execution agent’s goal mode, enter the /goal command, paste the entire task brief, and send it. No need to save it to a file or run any other startup command—paste once for each goal. After it finishes, let the author know. The author will rerun the acceptance commands and give you a brief report covering whether it passed, what was completed, what remains, and what to do next.

Why use a task brief instead of writing a prompt directly?

Handwritten prompts tend to run into three problems: first, commands are based on imagination, so when the agent encounters a nonexistent command, it can only fabricate one or blame the environment; second, the objective function may be wrong—for example, “make the tests pass” can encourage the agent to skip tests rather than fix the code; third, there is no way to resume if the process is interrupted. leader addresses each of these issues by testing every command first, designing acceptance checks that make cheating useless, and including a progress file to support resuming.

Is there a character limit for task briefs?

Yes. The 4,000-character limit is strict because it is the input limit for goal mode. If the task cannot fit, leader will recommend splitting it into several independent tasks and delivering one at a time instead of forcing everything into a single brief.