Results for “context-engineering”

Runtime router for context decisions during a task — decide read-more / stop / ask / assume / delegate, pick a minimal read-set, and time compaction. Front-page mental model is Write / Select / Compress / Isolate; a lazy pattern index routes to 25 context-engineering patterns, one reference per decision. Use when context is near budget, a long session is degrading, you must choose what to read for a task, or you are deciding /compact vs /clear vs a sub-agent. NOT for structural .claude/ overhead audits (see mk:context-audit); NOT for owning persistence (see mk:memory); NOT for USD cost (see mk:budget).

The English summary is being prepared. Showing the original description.

Learn how instructions, documents, tool results, and conversation history form an AI model’s context. Understand attention limits, why information in the middle can be missed, and how relevance and placement inform decisions about what to include.

• Learning the basics of context engineering
• Understanding why more input does not always help
• Explaining how instruction placement affects attention

Plan and improve AI agent systems by managing the information models receive. This collection covers memory, agent coordination, tool design, context compression, evaluation, and the operating controls needed for reliable autonomous workflows.

• Planning information management for a new AI agent
• Investigating failures in long agent conversations
• Designing agent memory and multi-agent coordination

Understand the components, mechanics, and constraints of context in agent systems. Use when writing, editing, or optimizing commands, skills, or sub-agents prompts.

The English summary is being prepared. Showing the original description.

Audits and reorganises what an AI coding agent carries in its context window: instruction files, path-scoped rules, skills, subagents, hooks and the running conversation. Use when someone says "the agent ignores our CLAUDE.md", "set this repo up for Claude Code / Codex / Cursor", "my AGENTS.md is too long", "quality drops in long sessions", "it keeps inventing APIs", "what should go in a skill versus the rules file", or asks how to hand work to a subagent. Grounded in the documented loading behaviour of Claude Code, with the equivalents for Codex, Gemini CLI and Cursor.

The English summary is being prepared. Showing the original description.

Expert en ingénierie du contexte IA (compression de prompts, chunking, gestion tokens, optimisation fenêtre)

The English summary is being prepared. Showing the original description.

Organize the information an AI coding agent needs for each task. This skill guides project rules, relevant file selection, conversation summaries, and session handoffs so the agent can follow existing conventions and retain key decisions.

• Set up project rules for AI coding agents
• Select relevant files before starting a task
• Trim stale information from a long conversation

Organize the instructions, conversation history, and tool information an agent receives. Use the skill to clarify business rules, structure SKILL.md files for loading when needed, and distinguish external source material from instructions to reduce prompt injection risks.

• Reorganizing an agent's system instructions
• Reviewing SKILL.md structure and activation conditions
• Turning vague business rules into explicit decisions

Give an agent rules for consulting long-term memory before work and saving decisions, corrections, and failures afterwards. It helps carry preferences and project conventions across sessions while preserving dated history and making conflicting memories visible.

• Carry preferences and project conventions across sessions
• Record user corrections for future tasks
• Preserve the history of changed decisions

Diagnose why an AI agent becomes less reliable as conversations and documents accumulate. Distinguish missed information, persistent false claims, irrelevant material, mixed tasks, and conflicting sources, then choose ways to clean, reorganize, or separate the context.

• Investigating declining quality in long conversations
• Recovering from false claims that persist after correction
• Resolving conflicts between retrieved documents

Turn ePub books into training examples for models that reproduce an author's writing style. Covers text extraction, coherent segmentation, varied instructions, model training, and checks for copied phrases and style transfer to new scenarios.

• Build training datasets from ePub books
• Train a model on an author's writing style
• Check for copied phrases in generated prose

Design evaluation systems that measure agent results across accuracy, completeness, citations, and tool efficiency. Build repeatable tests, compare changes against a baseline, and define quality checks that catch regressions before deployment and monitor production performance.

• Compare agent configurations against a baseline
• Catch quality regressions before deployment
• Measure the effects of context strategies

Organize content, contacts, knowledge, and goals in structured files. Use recorded writing preferences and past posts to draft content, prepare meeting briefs from relationship history, and review weekly progress.

• Drafting posts in a consistent personal voice
• Preparing meeting briefs from contact history
• Reviewing weekly progress and goals

Design infrastructure for agents that run in isolated remote environments. Plan faster session startup, saved work states, shared sessions, authentication, and delivery of code changes and other results back to users.

• Designing infrastructure for background coding agents
• Reducing delays when agent sessions start
• Saving and restoring session work

Design tools that agents can select and call correctly. Clarify descriptions, input rules, response formats, and recovery messages, then review overlapping tools and naming conventions using examples of successful and failed calls.

• Writing descriptions and input rules for new tools
• Investigating incorrect tool selection and calls
• Improving error messages with recovery steps

Plan projects that use large language models, from checking whether the task suits a model to designing processing stages. Estimate costs, choose single or multiple agents, and organize prototypes, structured outputs, and iteration.

• Checking whether a task is suitable for an LLM
• Designing a pipeline for batch analysis
• Estimating processing costs before development

Design loops that improve an agent’s own prompts, workflows, or execution code from failure records. Define bounded edits, preserve candidate histories, and accept changes through independent tests that check improvements and regressions.

• Designing agent self-improvement from failure records
• Defining acceptance checks for self-modifying workflows
• Diagnosing score gaming and stalled improvement

Use a structured starting point to write new context engineering skills. Organize activation conditions, responsibilities, practical steps, examples, and common mistakes, while keeping detailed references separate and clarifying boundaries with related skills.

• Draft a new SKILL.md
• Clarify when a skill should activate
• Define boundaries with related skills

Design systems where multiple agents work with separate task information. Choose coordination structures, define handoffs and shared state, and plan parallel work while checking whether communication costs and failure risks justify using more than one agent.

• Split large tasks across separate agent contexts
• Run independent research tasks in parallel
• Define handoffs between specialized agents

Model what an agent believes, wants, and commits to doing. Connect external facts to goals and plans using structured relationships, with evidence and validity periods that help trace why the agent chose an action.

• Convert RDF data into agent beliefs
• Trace the reasons behind agent decisions
• Track belief validity and changes over time

Design systems that use language models to judge AI responses. Covers scoring criteria, comparisons between two answers, bias reduction, confidence calibration, and checks against human judgments to make automated quality assessments more consistent and traceable.

• Compare responses from different models
• Run A/B tests for prompt changes
• Create shared scoring rubrics

Design the controls around autonomous agents: what they can edit, how results are evaluated, how progress is logged, and when human approval is required. Build research and experiment loops with clear recovery and review rules.

• Designing autonomous research and experiment loops
• Separating editable work from fixed evaluation rules
• Preserving progress and failed attempts across long runs

Explore how coordinating agents can pass relevant working state to workers through KV cache, a model's internal attention memory. Covers task-specific selection, runtime requirements, and comparisons of accuracy, token use, and processing overhead.

• Reducing repeated history sent to worker agents
• Comparing internal memory sharing with text summaries
• Assessing runtime support for KV cache compaction

Design how an AI agent edits its own working conversation history, choosing what to retain, rewrite, or remove. Covers safeguards, capacity monitoring, recovery after overflow, and the computing cost of changing earlier context.

• Designing agents that edit their own conversation history
• Preserving exact values during long tasks
• Recovering from context overflow