Resonant-Jones/The-Continuity-Atlas

Documentation on Best Practices of AI Assisted Coding

Summary

The Continuity Atlas is a documentation repository that lays out a methodology for working with AI coding agents, framed as a "field manual" rather than a runnable library. Its premise is that AI-assisted engineering is a continuity problem as much as a prompting problem, so it focuses on the surrounding structure: task design, specifications, validation, receipts, and drift detection. The repository is published under the MIT license.

What it is useful for

The Atlas targets developers using coding agents such as Codex, Cursor, Copilot, and Claude Code, alongside repo-aware assistants, local LLM tooling, multi-agent workflows, and hybrid local/cloud setups. It is most useful when a team needs evidence rather than confidence — bounded implementation tasks, separated planning and execution, recorded architectural decisions, and a way to reconstruct project state after context has fragmented.

The README identifies recurring situations where the Atlas applies: a repository that feels scattered or has been touched by multiple agents, documentation and implementation that may have diverged, a single person holding all project context, a current-state read needed before code changes, and small patches quietly turning into architectural changes. It also introduces the "Continuity Steward," a repo-local process companion intended to surface what is implemented, what is only planned, what has proof, and what is drifting.

How engineers can use it

The README documents a seven-step Quickstart:

1. Read docs/00-start-here.md.
2. Use templates/coding-agent-task.md during the next coding-agent session.
3. Walk through the examples/tiny-cli-refactor toy example.
4. Write a receipt with receipts/template.md.
5. Record architectural decisions with adr/template.md.
6. Run a drift review with templates/documentation-drift-review.md.
7. Use the Promptnomicon Steward at docs/promptnomicon-steward.md for repo-level reality checks.

The repo also ships templates for architecture planning prompts, proof-surface checklists, validation checklists, research-to-spec packets, campaign/spec directories, steward sessions, and a project-reality footer. Because this is a methodology repository rather than a runtime tool, there are no install commands; engineers consume it by reading the docs and adapting the templates into their own projects.