acaprino/codebase-mapper

Codebase Mapper Knowledge Base

Purpose

Generate a human-readable project guide for unfamiliar codebases. Output is narrative, didactic material - not technical dumps or AI-oriented analysis. Target audience: a smart colleague on their first day.

Output Structure

All output goes to .codebase-map/ in the project root:

.codebase-map/
  INDEX.md                    # Entry point with navigable summary
  00-executive-summary.md     # Plain-language summary for anyone (non-technical)
  01-overview.md              # What is this project, who is it for
  02-features.md              # Functional capabilities
  03-tech-stack.md            # Technologies and dependencies
  04-architecture.md          # How code is organized, layers, components
  05-workflows.md             # Main user/system flows with diagrams
  06-data-model.md            # Data structures, entities, relationships
  07-getting-started.md       # Where to start working, key files, dev setup
  08-open-questions.md        # Gaps, unknowns, things to ask the team
  09-project-anatomy.md       # Config files, env vars, scripts, directory tree
  10-configuration-guide.md   # Configuration recipes, operations, troubleshooting
  11-glossary.md              # Domain and technical glossary (plain definitions)
  _internal/
    context-brief.md          # Phase 1 exploration output (internal reference)
    interconnect.md           # Phase 1b structured map: contracts, invariants, domain rules (optional)

Core Principles

Tone

• Narrative, conversational, didactic
• Write for the audience in the Project Profile (see references/audience-adaptation.md); default to a smart colleague on their first day when no profile exists
• Progressive disclosure: big picture first, then details
• Honest about gaps - never fabricate or speculate

Content Rules

• Every technical term gets a brief inline explanation on first use
• File paths always included - reference actual code paths for every claim
• Diagrams inline - Mermaid blocks embedded in documents, not separate files
• No AI boilerplate: no "In this document we will...", no "Let's dive in", no trailing summaries

Diagram Standards

• Use Mermaid syntax exclusively
• Keep diagrams focused - max 15-20 nodes per diagram
• Split complex systems into multiple smaller diagrams
• Use descriptive node labels, not abbreviations
• Supported types: mindmap, flowchart, sequence, erDiagram, block-beta

Agent Coordination

Phase 1 - Explore

Single codebase-explorer agent reads the project and writes _internal/context-brief.md, which leads with a ## Project Profile (type, audience, register) and a ## Why / Context dossier.

Phase 1.5 - Confirm Profile

The map-codebase command surfaces the inferred Project Profile and lets the user confirm or adjust it before the writers run. The confirmed profile drives register and depth across all documents.

Phase 1b - Interconnect Map

Single senior-review:semantic-interconnect-mapper agent reads the context brief and writes _internal/interconnect.md (contracts, invariants, domain rules, integration hot-spots). Optional: if it fails, the pipeline continues in degraded mode. The tech-writer, flow-writer, ops-writer, and guide-reviewer cite this map instead of paraphrasing code.

Phase 2 - Write

Six parallel writer agents, each reading context-brief.md:

• overview-writer - 00-executive-summary.md (plain-language), 01-overview.md, 02-features.md (mindmap)
• tech-writer - 03-tech-stack.md, 04-architecture.md (component diagram)
• flow-writer - 05-workflows.md, 06-data-model.md (flowcharts, sequence, ER)
• onboarding-writer - 07-getting-started.md, 08-open-questions.md
• ops-writer - 09-project-anatomy.md (config files, env vars, scripts, directory tree)
• config-writer - 10-configuration-guide.md (config recipes, operations, troubleshooting)

Phase 3 - Review

Single guide-reviewer agent reads all documents, adds cross-references, fixes consistency, checks register consistency against the Project Profile, writes 11-glossary.md, and produces INDEX.md with per-audience reading paths.

References

Read on demand, not upfront:

• references/writing-guidelines.md - voice, tone, structure, audience as a parameter
• references/audience-adaptation.md - Project Profile schema, register matrix, archetypes, the "for everyone" rule
• references/diagram-patterns.md - Mermaid templates for each document

Standalone Documentation

Beyond the pipeline, the plugin provides standalone documentation agents:

• documentation-engineer - Bottom-up technical documentation from code analysis (API docs, architecture, tutorials, refactoring)
• doc-humanizer - Rewrites existing documentation to follow the writing guidelines

Both agents use the same writing guidelines and diagram patterns as the pipeline writers.