Interactive Tool

Project Foundry

A guided builder for creating a stronger AI project structure with contracts, handoffs, review rules, and durable corrections.

Framework nameOwner or teamProject namePrimary work typeOne-sentence missionTypical tasks

Generated Kit

Project Foundry

New Project • Mixed • Skeptical

Always contractsSkeptical reviewAggressive resetsScripts included

Folder structure

project-name/
  AGENTS.md
  README.md
  docs/
    identity.md
    current-work.md
    decisions.md
    terminology.md
    corrections.md
    agent-failures.md
  templates/
    contract-template.md
    handoff-template.md
    evaluator-template.md
  scripts/
    new-contract
    new-handoff
    new-review
    log-failure

AGENTS.md

# Project Foundry

This project uses a harness-first workflow for this project. Treat repeated agent mistakes as environment problems to solve with better structure, clearer docs, or stronger checks.

Before non-trivial work:
- Read only the docs relevant to the task.
- Create a contract before starting meaningful work.
- Prefer fresh-context restarts early rather than dragging long threads forward.

Core references:
- docs/identity.md
- docs/current-work.md
- docs/decisions.md
- docs/terminology.md
- docs/corrections.md
- docs/agent-failures.md

Workflow templates:
- templates/contract-template.md
- templates/handoff-template.md
- templates/evaluator-template.md

When a correction happens twice, encode it into the environment rather than repeating it in chat.

docs/decisions.md

# Decisions

## Mission
- Build a reusable environment for high-quality AI-assisted work.

## Locked Decisions
- Keep top-level instructions short and directional
- Use focused docs as the system of record
- Require explicit contracts for meaningful work

## Quality Rules
- State assumptions explicitly
- Judge work against the contract
- Log repeated failures

## Review Mode
- Skeptical

## Context Reset Policy
- Aggressive

## Handoff Trigger
- Create a handoff when quality drifts, context shifts, or the work pauses.

Bootstrap prompt

You are helping set up this project using the Project Foundry harness.

Operator:
- Project owner not specified

Mission:
- Build a reusable environment for high-quality AI-assisted work.

Work type:
- Mixed

Typical tasks:
- Draft recurring work with less re-explaining
- Review output against explicit success criteria
- Capture failures and corrections for future runs

Terminology:
- Define any project-specific language here

Voice:
- Direct
- Warm
- Concise
- Implementation-oriented

Anti-patterns:
- Generic AI filler
- Vague praise without evaluation
- Long answers when short ones are enough

Corrections:
- Add repeated corrections here as durable rules

Review policy:
- Use a skeptical evaluator and fail work that does not fully meet the contract.

Failure logging:
- Log repeated misses, skipped verification, bad terminology, and premature done calls.

Do not create one giant instruction blob. Keep the top-level map short and put durable truths into focused docs.