What is true now?
An authoritative current state covering requirements, architecture, implementation status, known issues, and next steps.
Downloadable Codex skill
maintain-project-document
The more capable AI becomes, the more demanding our responsibility becomes: give it complete context.
AI models can now undertake increasingly complex, end-to-end implementation tasks. But intelligence alone isn’t enough. The AI still needs a thorough account of what we are building, why we are building it, the requirements and constraints, architecture and design decisions, previous attempts, known issues, uncertainties, and next steps.
That context rarely exists in one person’s memory or a single conversation. It develops progressively across meetings, planning sessions, investigations, experiments, and earlier AI sessions. Important details need to be recorded safely on disk, refined over time, and kept internally consistent before the project is ready to be handed to AI for implementation.
I created a Codex skill called maintain-project-document to support this process and keep the project record aligned with the work as it evolves.
Two distinct views
How did we get here?
A concise chronological activity log recording decisions, changes, tests, and outcomes.
The skill also
- Uses a confirmed Jira ticket or project identifier
- Finds and updates existing documents instead of creating duplicates
- Maintains project indexes automatically
- Keeps evidence organized and sanitized
- Checks for contradictions and obsolete information before finishing
The key design decision was separating “What is true now?” from “How did we get here?”
That makes the document useful to someone joining the project today—without forcing them to reconstruct the current state from weeks of meeting notes, chat threads, and commit history.
Good documentation shouldn’t be a snapshot.
It should be part of the project workflow.
The complete package includes SKILL.md, configuration metadata, and the project-document template.
Download skill (.zip)