Agent-Skills.md

Agent Skills: Document Project Task

Use to analyze and document any project codebase. Creates comprehensive reference documentation for AI-assisted development including architecture and patterns.

UncategorizedID: resolve-io/.prism/document-project

Repository

resolve-io/.prism
resolve-io
12

Install this agent skill to your local

pnpm dlx add-skill https://github.com/resolve-io/.prism/tree/HEAD/plugins/prism-devtools/skills/document-project

Skill Files

Browse the full folder contents for document-project.

Download Skill

Loading file tree…

plugins/prism-devtools/skills/document-project/SKILL.md

Skill Metadata

Name
document-project
Description
Use to analyze and document any project codebase. Creates comprehensive reference documentation for AI-assisted development including architecture and patterns.

Document Project Task

When to Use

  • Project Documentation: Document any codebase for AI understanding
  • Architecture Assessment: Analyze current system architecture
  • Knowledge Transfer: Create comprehensive project documentation
  • PRD Preparation: Generate context needed for PRD creation
  • Onboarding: Build complete project reference for new team members

What This Skill Does

Guides you through building comprehensive project documentation:

  • Configuration: Load project settings from .prism/core-config.yaml
  • Classification: Detect project type, structure, and technology stack
  • Scanning: Analyze codebase (quick, deep, or exhaustive)
  • Documentation: Generate architecture, API, and development guides
  • Indexing: Create master index for AI retrieval

Quick Start

  1. Run the skill and specify project root directory
  2. Choose scan depth (quick/deep/exhaustive)
  3. Confirm project classification
  4. Wait for documentation generation
  5. Review generated files at {{output_folder}}/index.md

Output Location: docs/project/ (or as configured)

Workflow Overview

The workflow follows 10 sequential steps. For detailed instructions, see Workflow Steps Reference.

| Step | Name | Purpose | |------|------|---------| | 0 | Load Configuration | Read .prism/core-config.yaml | | 1 | Check Existing | Detect existing docs, choose mode | | 2 | Select Scan Level | Quick/Deep/Exhaustive | | 3 | Initialize State | Create tracking file | | 4 | Detect Structure | Classify project type | | 5 | Scan Project | Execute selected scan depth | | 6 | Deep-Dive Mode | (If selected) Detailed area analysis | | 6.5 | Check Existing Docs | Avoid duplicates via Smart Connections | | 7 | Generate Index | Create master entry point | | 8-10 | Validate & Complete | Verify, fill gaps, finalize |

Scan Levels

| Level | Time | What It Does | |-------|------|--------------| | Quick | 2-5 min | Pattern-based analysis, configs only | | Deep | 10-30 min | Critical directories, selective file reading | | Exhaustive | 30-120 min | ALL source files (excludes node_modules) |

Project Types Detected

  • claude-code-plugin: PRISM-style plugins with skills/commands/tasks
  • web: React, Vue, Angular applications
  • backend: Express, FastAPI, Rails servers
  • mobile: React Native, Flutter apps
  • cli/library/desktop: Various other project types

Generated Documentation

Core files generated (as applicable):

  • index.md - Master entry point
  • project-overview.md - Executive summary
  • architecture.md - Technical architecture
  • technology-stack.md - Complete tech stack
  • development-guide.md - Setup and workflow
  • api-contracts.md - API endpoints
  • data-models.md - Database schema
  • source-tree-analysis.md - Directory structure

Key Principles

  1. Write-as-you-go: Write files immediately, don't accumulate
  2. Resumability: State file tracks progress for resume
  3. Batching: Process subfolders one at a time for token management
  4. No duplicates: Check existing docs before creating

Reference Documentation

Guardrails

  • Always load config first (Step 0)
  • Update state file after EVERY step completion
  • Write documents immediately after generating content
  • Purge detailed data from context after writing
  • First-time startup may require explicit scan level selection