tldr-expert_skill

Overview

This skill is TLDR Expert: a Graph-Assisted Code Architect that delivers full repository comprehension while minimizing token cost. It focuses on semantic mapping, automated context packing (ACP), and token-efficient digests so agents can reason over large codebases with less context overhead. The skill turns repos into prompt-ready context bundles and maintains a cross-file dependency graph for precise tracing.

How this skill works

TLDR Expert builds a semantic index and call graph (llm-tldr) and produces compressed, signature-only context bundles with Repomix and Gitingest. It extracts function/type signatures via Tree-sitter, creates digestible summaries, and attaches dependency traces so agents see high-fidelity structure without full implementation noise. The pipeline includes index warming, semantic search, targeted context packaging, and secretlint checks to avoid leaking sensitive data.

When to use it

• Onboarding new engineers or sub-agents to a repo quickly
• Preparing focused context for long-context models (o3, Gemini 3) before planning
• Tracing cross-file call paths and dynamic dependencies
• Reducing prompt token usage when exploring large monorepos
• Creating audit-safe, prompt-ready digests for review or automation

Best practices

• Always run tldr status and tldr warm . after large refactors to keep the index fresh
• Package only relevant subdirectories with --signatures-only or --top-level-only to avoid bloat
• Combine semantic search with callers/callees tracing rather than relying on grep for dependency discovery
• Use Repomix ignore rules to exclude node_modules, dist, and secrets from bundles
• Validate semantic search hits by opening the few most relevant signature blocks before relying on them for decision-making

Example use cases

• Create a Repomix context bundle for the auth feature and pass it to the reasoning model for a secure redesign plan
• Use llm-tldr to find all callers of session management code, then pack a minimal context for a refactor ticket
• Generate a Gitingest onboarding digest for a new hire to reach 80% domain familiarity in minutes
• Warm the semantic index after CI changes and run targeted ACP before an automated migration agent executes