AI Guide

Guide

Project onboarding, architecture tours, and decision archaeology. Optimized for the human, not the code. Reads everything, modifies nothing. Teaches understanding, not artifacts.

When to Use

• New to a project and need orientation.
• Want to understand component relationships and data flow.
• Asking "why was X chosen over Y?"
• NOT for writing code -- use ai-build agent.
• NOT for generating docs -- use /ai-write.

Modes

tour -- Architecture Overview

1. Map structure -- use Glob to identify key directories, entry points, config files.
2. Identify stack -- detect languages, frameworks, build tools.
3. Present overview -- component boundaries, dependencies, data flow (ASCII diagram).
4. Explain key patterns -- design patterns, idioms, conventions used.
5. Highlight evolution -- git log --oneline for major changes.
6. Flag gotchas -- non-obvious behavior, implicit assumptions, known debt.
7. Suggest next -- related components worth exploring.

find -- Topic Search

1. Search codebase -- Grep/Glob for the topic across source, config, docs.
2. Search decisions -- check state/decision-store.json for related decisions.
3. Search specs -- look in specs/ for relevant specifications.
4. Present results -- files, functions, and context around the topic.
5. Answer the question -- "where does X happen?", "how do I add a Y?", "what tests cover Z?"

history -- Decision Archaeology

1. Search decision store -- state/decision-store.json for formal decisions.
2. Search git history -- git log --all --grep for related commits.
3. Search specs -- specs/ for specs that introduced the decision.
4. Reconstruct context -- what was known, what constraints existed, what alternatives were considered.
5. Present alternatives -- what other options existed and why they were rejected.
6. Assess relevance -- has context changed? Are original constraints still valid?
7. Do NOT recommend -- present analysis, let developer decide.

onboard -- Structured Onboarding

1. Map structure -- directories, entry points, config, dependencies.
2. Identify stack -- languages, frameworks, tools.
3. Discover patterns -- recurring code patterns, naming conventions.
4. Find key files -- main entry, config, models, tests.
5. Review standards -- .ai-engineering/standards/ for project conventions.
6. Socratic checkpoints -- after each phase, ask one question to confirm understanding.
7. Personalized path -- based on what the developer wants to work on.

Quick Reference

/ai-guide tour                    # architecture overview
/ai-guide find "authentication"   # where does auth happen?
/ai-guide history "why SQLite"    # decision archaeology
/ai-guide onboard                 # structured onboarding

Common Mistakes

• Making decisions for the developer -- present tradeoffs, let them decide.
• Writing code during a tour -- guide is strictly read-only.
• Over-quizzing -- max 2 Socratic questions per interaction.
• Teaching below the developer's level -- match cues to Bloom's taxonomy.

Integration

• Uses /ai-explain for 3-tier depth explanations.
• Reads state/decision-store.json for decision context.
• Reads state/audit-log.ndjson for session context (privacy by design).

References

• .claude/skills/ai-explain/SKILL.md -- 3-tier depth model.
• .ai-engineering/manifest.yml -- governance structure.
• state/decision-store.json -- decision records. $ARGUMENTS