Tabletop Tracker
Start here
| If you want to… | Read |
|---|---|
| Understand the whole thing | Architecture overview |
| Know why something is the way it is | Decision catalogue — every ADR, numbered and dated |
| Contribute, human or agent | The loop |
| See what is being built next | Work backlog |
How this project is built
Work is done by AI agents running a defined loop, with a human at the merge.
One work item is one issue, one branch, one pull request. Agents loop freely on a branch — spec, failing test, implement, verify, document, self-review, open a PR carrying evidence — and never merge. The gate decides whether something is done, not the agent’s own assessment.
The failure mode the whole system is designed against is confident false completion: an agent reporting success it did not achieve. Every mechanism exists to make “done” machine-checkable rather than self-assessed.
About these docs
This site is generated from the docs/ directory of the repository, with no copy step and no duplicated files. Editing a decision record updates this site in the same commit — see ADR-0013.