Project Wiki AI Skills

Project Wiki AI Skills represent a suite of autonomous capabilities designed to transform a codebase into research-grade documentation. Unlike standard documentation generators that simply summarize files, these skills utilize a research-type Agent approach to explore design intent, architectural evolution, and complex data flows. The resulting Wiki is stored locally in .atmos/wiki/ as a set of structured Markdown files and a JSON-based catalog.

Sources: skills/project-wiki/SKILL.md:1-15, skills/project-wiki/README.md:1-10

Core Skill Variants

The project defines three primary AI skills tailored for different stages of the documentation lifecycle:

Sources: skills/project-wiki/SKILL.md:5-10, skills/project-wiki-update/SKILL.md:5-10, skills/project-wiki-specify/SKILL.md:5-10

Research and Generation Workflow

The AI Skill follows a strict, multi-step process to ensure technical accuracy and depth. The "Research-Type" role differentiates it from document summarizers by requiring the agent to answer specific research questions and trace data flows through entry points, core types, and implementation files.

Metadata Collection and Analysis

Before writing, the agent must collect Git metadata (commit logs, PRs, and issues) to understand why the code evolved. This information is stored in .atmos/wiki/_metadata/ and informs the creation of _concepts.json, which maps recurring design patterns and technical decisions.

Sources: skills/project-wiki/SKILL.md:50-70, skills/project-wiki/SKILL.md:105-125

Generation Sequence

The following diagram illustrates the sequence from initial survey to the final validation of generated content.

Incremental Update Strategy

The project-wiki-update skill employs a three-phase hybrid strategy to maintain documentation sync without full regeneration. This ensures that even "uncovered" files (those not previously documented) are analyzed for impact.

1. Phase 1: Source-Matched Pages: Identifies pages whose sources frontmatter matches files in git diff <old>..HEAD.
2. Phase 2: Uncovered Changes Analysis: AI reasons if new files require new pages, updates to existing pages, or can be skipped.
3. Phase 3: Cross-Cutting Impact Check: Checks if shared utilities or core abstractions have changed, affecting documentation logic even without direct file references.

Sources: skills/project-wiki-update/SKILL.md:25-45, specs/tech/wiki_incremental_update.md:250-280

Content Depth and Quality Standards

All skills enforce a strict set of metrics to ensure the Wiki is useful for maintainers and contributors. These are verified by automated validation scripts (validate_content.py).

Sources: skills/project-wiki/SKILL.md:225-240, skills/project-wiki-specify/SKILL.md:45-55

Architecture and Data Structure

The Wiki is governed by a _catalog.json file which defines a hierarchical navigation tree. Each Markdown document contains YAML frontmatter used by the frontend for rendering and metadata display.

Catalog Schema

The catalog must include a commit_hash to support incremental updates and a nested project object.

{
  "version": "2.0",
  "generated_at": "2026-02-11T12:00:00Z",
  "commit_hash": "abc1234",
  "project": {
    "name": "ATMOS",
    "description": "Project Description"
  },
  "catalog": []
}

Sources: skills/project-wiki/SKILL.md:150-180, specs/tech/wiki_incremental_update.md:40-50

Markdown Frontmatter

Metadata is strictly YAML-based to allow for reliable frontend parsing.

---
title: WebSocket Service Architecture
section: deep-dive
level: advanced
reading_time: 13
path: deep-dive/infra/websocket
sources:
  - crates/infra/src/websocket/manager.rs
  - apps/api/src/api/ws/handlers.rs
updated_at: 2026-02-10T12:00:00Z
---

Sources: skills/project-wiki/SKILL.md:245-265, skills/project-wiki/references/output_structure.md:120-140

Integration with ATMOS UI

The AI skills are integrated into the ATMOS frontend through the "Wiki Tab" in CenterStage. The system detects the presence of .atmos/wiki/ and routes the user either to a setup flow (invoking the project-wiki skill) or a viewer flow.

Project Wiki AI Skills provide a self-healing and evolving documentation system. By combining deep codebase research with Git-based change detection, they ensure that technical documentation remains an accurate, high-fidelity representation of the underlying software architecture.