Help us improve
Share bugs, ideas, or general feedback.
From context-engineering
Principles for designing context-efficient AI agents and tools. Use when designing LLM tools, agents, MCP servers, or multi-agent systems.
npx claudepluginhub byunk/skills --plugin context-engineeringHow this skill is triggered — by the user, by Claude, or both
Slash command
/context-engineering:context-engineeringThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Principles and patterns for strategically designing the information provided to LLMs.
references/01-quality-over-quantity.mdreferences/02-start-minimal.mdreferences/03-hierarchical-organization.mdreferences/04-multi-dimensional-context.mdreferences/05-simple-composable-patterns.mdreferences/06-invest-in-tool-design.mdreferences/07-match-architecture-to-complexity.mdreferences/08-transparency-in-planning.mdreferences/09-contextualize-retrieved-chunks.mdreferences/10-dynamic-context-adaptation.mdreferences/11-meta-context.mdreferences/12-ground-in-environment.mdreferences/13-test-and-measure.mdreferences/14-retrieve-generously.mdreferences/15-anti-patterns.mdreferences/16-guide-model-thinking.mdreferences/17-multi-window-state.mdreferences/18-calibrate-autonomy.mdMandates invoking relevant skills via tools before any response in coding sessions. Covers access, priorities, and adaptations for Claude Code, Copilot CLI, Gemini CLI.
Share bugs, ideas, or general feedback.
Principles and patterns for strategically designing the information provided to LLMs. Context is a critical but finite resource — these principles help you use it well.
| # | Principle | File | Summary |
|---|---|---|---|
| 1 | Quality Over Quantity | 01-quality-over-quantity.md | Include only relevant context. Precision beats exhaustiveness. |
| 2 | Start Minimal, Layer Progressively | 02-start-minimal.md | Begin with essentials, expand based on performance gaps. |
| 3 | Hierarchical Context Organization | 03-hierarchical-organization.md | Structure context from general to specific. Use XML tags for disambiguation. |
| 4 | Multi-Dimensional Context | 04-multi-dimensional-context.md | Effective context spans task, domain, history, constraints, motivation, and format. |
| 5 | Simple Composable Patterns | 05-simple-composable-patterns.md | Prefer prompt chaining, routing, parallelization over complex frameworks. |
| 6 | Invest in Tool Design (ACI) | 06-invest-in-tool-design.md | Tool descriptions deserve as much effort as UIs. Calibrate action language for Claude 4.6. |
| 7 | Match Architecture to Complexity | 07-match-architecture-to-complexity.md | Choose the right pattern: chaining, routing, parallelization, orchestrator, or agent. |
| 8 | Transparency in Agent Planning | 08-transparency-in-planning.md | Show planning steps. Use structured state tracking for long-horizon tasks. |
| 9 | Contextualize Retrieved Chunks | 09-contextualize-retrieved-chunks.md | Prepend explanatory context before embedding. Reduces retrieval failures by 35-67%. |
| 10 | Dynamic Context Adaptation | 10-dynamic-context-adaptation.md | Adjust context based on task complexity and model confidence. |
| 11 | Meta-Context | 11-meta-context.md | Explain why information is provided. Use grounding directives to reduce hallucinations. |
| 12 | Ground Agents in Environment Feedback | 12-ground-in-environment.md | Use tool results and human checkpoints, not just model reasoning. |
| 13 | Test and Measure Systematically | 13-test-and-measure.md | A/B test context variations. Measure accuracy, latency, reasoning quality. |
| 14 | Retrieve Generously, Filter Late | 14-retrieve-generously.md | Top-20 chunks outperform top-5. Don't over-filter prematurely. |
| 15 | Avoid Common Anti-Patterns | 15-anti-patterns.md | Don't dump everything, over-abstract, overengineer, or inherit stale aggressive prompts. |
| 16 | Guide Model Thinking | 16-guide-model-thinking.md | Control reasoning depth with adaptive thinking. Prevent overthinking on simple tasks. |
| 17 | Multi-Window State Management | 17-multi-window-state.md | Externalize state for tasks spanning multiple context windows. |
| 18 | Calibrate Autonomy and Safety | 18-calibrate-autonomy.md | Tune action thresholds. Claude 4.6 defaults to high autonomy; set guardrails. |