qshanx/docs-governance

What it solves

When AI helps write code, the code becomes cheap to regenerate, but the reasoning behind it—why a decision was made, what tests prove, which areas are risky—gets lost. Each new session or agent can end up rediscovering the same structure, while READMEs go stale and architecture rationale disappears. docs-governance treats project documentation as a small system that preserves confirmed context, decisions, and handoff evidence so good work survives the session and becomes a reusable project capability.

How it works

Instead of forcing you to write more documents, the plugin sets up a thin documentation spine with four files, each with a single responsibility: durable rules (CLAUDE.md), navigation and do-not-touch areas (CLAUDE_MAP.md), current health and priorities (PROJECT_STATUS.md), and an append-only project history (PROJECT_LOG.md). It reuses existing sources where possible and keeps existing PRD/Spec files authoritative. The system works through a set of focused skills that you can invoke from Claude Code, Codex, or ChatGPT—for example, auditing, architecture docs, daily maintenance, or writing for agents. Audits are read-only by default; setup does not implicitly install hooks, initialize Git, commit, or push. The system uses Markdown as the source of truth, with an optional rebuildable SQLite index that is gitignored.

Who it’s for

This is for teams and individuals running long-term AI-assisted projects with tools like Claude Code, Codex, or ChatGPT. It’s especially useful when multiple agents or sessions need to share a consistent understanding of project rules, decisions, and verification evidence—and when you want to avoid losing context between sessions.

Highlights

• Read-only audits by default: /governance-audit produces a report, not an automatic rewrite.
• Thin documentation spine with clear single responsibilities for each file.
• Reuses existing PRD/Spec files and existing navigation; extensions stay optional.
• Optional SQLite index is rebuildable and gitignored; Markdown remains the source of truth.
• Includes focused skills for architecture docs, change impact, contract-first collaboration, and module regression.
• Written to be agent-friendly: progressive disclosure, completion criteria, and a local final check for AI-read documents.