Single-repo operating pack guide
AI Project Pack
A public single-repo operating pack for AI-assisted development. Keep current truth, decisions, pending items, next actions, sources, and update logs in one AI-readable repository instead of rebuilding project context every session.
Current status
Public v1 is live The repository is published and intended for immediate first-use, not as a draft or placeholder.Core loop
Report-only → safe-update → update log The default path keeps edits bounded and leaves a visible restart trail for the next model or session.Repository shape
Single AI-readable repo Current truth, decisions, pending items, next actions, sources, examples, playbooks, adapters, and append-only updates live together.Model styles
ChatGPT / Claude / Codex The repository includes adapter guidance for ambiguity reduction, evidence hygiene, and bounded repo edits.Support future maintenance and examples
If this page or repository is useful, support helps fund future examples, validation improvements, and documentation upkeep.
What it is
AI Project Pack is a public, single-repository operating pack for AI-assisted development. It gives a project one durable operating surface so a new AI session does not have to guess what is true, what is unresolved, and what should happen next.
The product is the repository itself. This page exists to explain that repository clearly before you open GitHub.
It is not a memory service, not an autonomous manager, and not a hosted platform. It is a reviewable text-first operating repo.
Why it exists
Projects drift when decisions stay trapped in chat, unresolved items get treated as settled, sources are not visible, and each new model session has to reconstruct context again from scratch.
AI Project Pack fixes that by separating the work into explicit, durable files:
- current truth
- decisions
- pending items
- next actions
- sources
- append-only update logs
Who it is for
Use it when project quality drops because context keeps resetting between AI sessions.
- product and technical planning with iterative AI support
- research-heavy implementation where evidence quality matters
- OSS release preparation with model handoff across sessions
- mixed human + AI teams that need auditable changes
It is a poor fit if you want autonomous project management or a platform that stores and runs work for you.
What is inside the repository
The repository is organized so AI can read, reason, and update within a visible operating surface instead of relying on hidden context.
Core operating files
current-truth.md, decisions.md, pending.md, next-actions.md, sources.md, and updates/.
Operational guidance
examples/, playbooks/, adapters/, docs/, templates/, and validation via scripts/validate_repo.py.
How it works
The repository is designed around one simple loop that stays readable and reviewable.
updates/.First-use flow
A first-time user should be able to do this in one loop:
- Open the repository.
- Read
README.mdandAGENTS.md. - Read
current-truth.md,decisions.md,pending.md,next-actions.md, andsources.md. - Run a report-only pass.
- Approve one bounded safe-update.
- Append one dated update log.
- Restart from the repo in the next session.
The examples inside the repo show this path explicitly, including report-only output, safe-update scope, touched-files reporting, and handoff instructions.
What makes it different
This is not just a prompt pack. It gives AI a durable operating surface with explicit read order, bounded update rules, source-aware claims, append-only history, and restart quality across sessions and models.
Future lightweight builder
A lightweight in-browser builder may be added later, but the repository remains the primary product.
If that lighter version ships, it should work as an entry point into the full repo, not as a replacement for the repository itself. The likely role would be a browser-based builder for project snapshot, decisions, pending items, next actions, sources, and AI prompts that then exports back into the repo workflow.
FAQ
Is AI Project Pack a website or SaaS?
No. The core product is the repository itself. This page explains how to use the repo clearly before opening GitHub.
Do I need Codex to use it?
No. The repository is useful with ChatGPT, Claude, Codex, and other AI tools that can read and work from repository files.
What is the default workflow?
The default workflow is report-only first, then one bounded safe-update, then an appended update log so the next session can restart safely.
Does it get better over time?
Yes, but only through disciplined use. The repository gets stronger when decisions, sources, unresolved items, and update logs remain coherent over repeated passes.
Will there be a lightweight site later?
Possibly. A lightweight builder may be added later, but the repository itself remains the primary product and operating surface.
Links
Support this project
If this page or repository helps your workflow, you can support future maintenance, validation improvements, and additional worked examples through the links below.
Disclaimer
This page and the linked repository are provided for informational and practical workflow use. They are offered as-is, without warranty of completeness, fitness for a particular purpose, or continued availability. Any use in real development or shipping workflows is at your own discretion. Contents may change over time.