# Be-Ambitious Harness

> A Claude Code skill and markdown workflow that uses a planner, worker and independent verifier to make AI agents tackle more ambitious tasks with a cheap gate check.

Be-Ambitious Harness is an open-source prompt engineering framework and multi-agent workflow published by alexpiar (Chernov Alexandr). It turns a task into a seven-section prompt with a baseline bar, an ambitious bar and a gate check, then routes execution through separate planner, worker and verifier roles. It ships as a Claude Code skill in English and Russian, and v0.1.0 is the first public release.

## What It Is

The harness is a folder of plain markdown files (agent instructions, core templates, scenario playbooks and checklists) rather than executable code. A human gives the planner a task, context, a mode (keep scope, ambitious extension or recommend) and an option to stop after analysis. The planner produces a short analysis, up to three questions and a prompt with seven fixed sections.

It installs by cloning the repository into the Claude Code skills directory and invoking `/be-ambitious-harness`. The README states that any agent able to read a folder of markdown files can use `SKILL.md` as an entry point.

## How the Workflow Operates

- **Two bars:** a baseline result and a more ambitious one, with the extra work and main uncertainty named. For unfamiliar tasks the ambitious bar is phrased as a learning goal.
- **Gate check:** a cheap, observable test with a pass criterion fixed before it runs and defined actions for three outcomes: passed, failed, insufficient data.
- **Role isolation:** the worker runs in a fresh session without the planner's analysis, and the verifier sees only requirements, acceptance criteria, the result and the evidence log, preferably on a different model.
- **Human ownership:** scope and budget decisions stay with the human, and a passed gate with no budget returns the baseline plus a cost estimate.

## Scenario Playbooks and Evals

Six scenario playbooks cover design/UX, data analysis, code engineering, writing/research, decision/plan and generic tasks, each with domain-specific gates and traps. An eval suite of eight tasks and a 30-criteria scorecard checks whether the harness itself holds up. The README reports that two tasks have been run so far, describes them as a smoke test rather than a benchmark, and notes that the single-skill comparison produced deeper results while the harness showed more discipline.

## Current Status

The README documents a known defect, that the verifier does not receive the gate criterion, with a fix pending. It also lists gaps: no multimodal scenario and no multi-gate support for multi-day tasks. The project states it is not affiliated with or endorsed by Anthropic.

## Features
- Two-bar task framing (baseline and ambitious)
- Gate check with pre-fixed pass criterion and three outcomes
- Planner, worker and verifier role isolation
- Seven-section prompt template
- Six scenario playbooks
- Eval suite with 8 tasks and 30-criteria scorecard
- Pre-flight and post-run checklists
- English and Russian files

## Integrations
Claude Code

## Platforms
API, CLI

## Pricing
Open Source

## Version
v0.1.0

## Links
- Website: https://github.com/alexpiar/be-ambitious-harness
- Documentation: https://github.com/alexpiar/be-ambitious-harness
- Repository: https://github.com/alexpiar/be-ambitious-harness
- EveryDev.ai: https://www.everydev.ai/tools/be-ambitious-harness