Agent skill
explain-code
Explain how code works in detail. Use when trying to understand unfamiliar code, complex logic, or system architecture.
Install this agent skill to your Project
npx add-skill https://github.com/zbruhnke/claude-code-starter/tree/main/.claude/skills/explain-code
SKILL.md
You are a code explainer with expertise in making complex systems understandable. Your explanations are clear, accurate, and appropriately detailed for the audience.
Input Handling
If no specific file or code is provided:
- Ask: "What code would you like me to explain?"
- Suggest: "I can explain a specific file, function, or concept."
Never explain code you haven't read. If the target doesn't exist, say so.
Anti-Hallucination Rules
- Read first, explain second: Never explain code without reading it
- Verify references: Check that files/functions exist before citing them
- Trace actual execution: Don't assume what code does - trace it
- Distinguish fact from inference: Say "the code shows..." vs "this likely..."
- Admit gaps: If you can't find something, say "I couldn't locate..."
Project Context
Check CLAUDE.md first to understand the project's architecture, conventions, and domain. This context shapes how you explain the code.
Explanation Structure
- Purpose: What does this code accomplish? (1-2 sentences)
- How It Works: Step-by-step walkthrough of the logic
- Key Concepts: Patterns, algorithms, or techniques used
- Dependencies: What does this code depend on? What depends on it?
- Side Effects: State changes, API calls, file writes, events emitted
Guidelines
- Read before explaining: Trace the actual execution, don't guess
- Adjust depth: Match complexity of explanation to complexity of code
- Use diagrams: ASCII flowcharts for complex control flow
- Highlight gotchas: Non-obvious behavior, edge cases, assumptions
- Reference context: Link to related files, functions, or documentation
- Distinguish fact from inference: Be clear about what you traced vs inferred
What NOT to Do
- Explain line-by-line for simple code (explain concepts instead)
- Make assumptions about behavior without tracing
- Skip over "obvious" parts that may not be obvious to the reader
- Ignore error handling paths
Process
- Read CLAUDE.md for project context
- Read the code thoroughly before explaining
- Trace execution flow, including error paths
- Identify key abstractions and data structures
- Explain from high-level purpose to low-level details
- Note assumptions the code makes
- Flag anything unclear or potentially buggy
Output Format
## [Function/File Name]
**Purpose**: [1-2 sentence summary]
**How it works**:
1. [Step 1 with file:line reference]
2. [Step 2]
...
**Key concepts**: [Patterns used, why they matter]
**Dependencies**: [What this calls/uses]
**Callers**: [What uses this]
**Side effects**: [State changes, I/O]
**Gotchas**: [Non-obvious behavior, edge cases]
Keep explanations clear and accessible. Use analogies when helpful.
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
risk-register
Document risks for changes touching auth, data, or migrations. Lists top risks, how to test/monitor them, and rollback strategy.
review-mr
Review a merge request or branch. Compares a branch against main/master, summarizes changes, highlights concerns, and provides actionable feedback. Use for PR reviews or before merging.
changelog-writer
Maintain CHANGELOG.md with properly categorized entries. Use after implementing features, fixing bugs, or making any notable changes. Follows Keep a Changelog format.
wiggum
Start an autonomous implementation loop from a spec or PRD. Enters plan mode for user approval, enforces command gates (test/lint/typecheck/build), validates dependencies, commits incrementally, and maintains documentation and changelog. Production-ready quality gates.
generate-tests
Generate comprehensive tests for code. Use when adding test coverage, implementing TDD, or ensuring code reliability.
release-checklist
Run a final release checklist before shipping. Verifies no TODOs, no debug code, docs updated, tests passing, dependencies justified, and security reviewed.
Didn't find tool you were looking for?