TLDR Expert
Overview
Achieves high-fidelity codebase comprehension at a fraction of the token cost through semantic layers, structured digests, and advanced context packaging. Combines Repomix for context packing, Gitingest for repository digests, and llm-tldr for graph-based code analysis.
When to use: Reducing prompt overhead for large codebases, onboarding to unfamiliar repositories, mapping cross-file dependencies, creating AI-optimized context bundles.
When NOT to use: Small single-file tasks, final implementation debugging (read the full file), real-time code editing.
Quick Reference
| Pattern | Tool / Command | Key Points |
|---|---|---|
| Context packing | repomix --include "src/**" --compress | Package subdirectories into AI-optimized bundles |
| Signatures only | repomix --include "src/**" --compress | Compression extracts signatures via Tree-sitter |
| Repository digest | gitingest . -o digest.txt | Prompt-friendly summary for quick onboarding |
| Dependency context | tldr context funcName --project . | LLM-ready context for a function with 95% token saving |
| Caller tracing | tldr impact functionName . | Reverse call graph to assess change blast radius |
| Forward call graph | tldr calls . | Build forward call graph across the project |
| Semantic search | tldr semantic "session expiry" . | Find logic by meaning when naming is inconsistent |
| Architecture audit | tldr arch . | Detect circular deps, layer violations, dead code |
| Dead code finder | tldr dead . | Find unreachable functions with zero callers |
| File extraction | tldr extract src/file.ts | Extract AST (functions, classes, imports) from a file |
| Secret scanning | Repomix built-in secretlint | Ensure context bundles contain no keys or PII |
Common Mistakes
| Mistake | Correct Pattern |
|---|---|
| Reading entire large files without checking structure first | Run tldr extract to get signatures before reading full files |
Using grep for dependency tracing across files | Use tldr impact for reverse call graph that understands dynamic imports |
Packing node_modules or dist into context bundles | Configure Repomix ignore-list to exclude generated and vendor directories |
| Assuming semantic search results are exhaustive | Verify top matches against actual source and cross-reference with rg |
| Running Repomix without compression on large directories | Use --compress flag to stay within context window limits |
| Including irrelevant context that dilutes signal quality | Follow top-down priority: index, signatures, core logic, then adjacent context |
Delegation
- Repository structure discovery: Use
Exploreagent to map directory layout and identify key modules before building context bundles - Multi-step context packing workflow: Use
Taskagent to run Gitingest digest, Repomix compression, and llm-tldr indexing in sequence - Architecture analysis and planning: Use
Planagent to design context engineering strategy for large monorepos
References
- Context Engineering Patterns -- packing strategies, XML tagging, signal-to-noise optimization, warm-up prompts
- Repomix and Gitingest Mastery -- configuration, compression mode, digest generation, Tree-sitter extraction
- Semantic Graph Analysis -- llm-tldr CLI tools, impact analysis, semantic search, architectural audits