Knowledge, Context and Agentic Engineering for Knowledge Work

Maintained knowledge and selected context

Module 03 · Knowledge and Context Engineering

The maintenance of reusable knowledge and the deliberate selection of task-specific context.

Scope and learning question

Learning question: Which knowledge should persist, which state belongs to the current task and which material should enter the working context now?

Knowledge Engineering creates and maintains explicit, inspectable and revisable project knowledge. Context Engineering selects, organises and provides the task-specific information state in which a prompt is interpreted.

Learning objectives

  • Distinguish a Project Knowledge Base, Working Context and technical context window.
  • Distil bounded Knowledge Documents while preserving relevant provenance, uncertainty and revision paths.
  • Select information, instructions, retrieval routes and tool outputs for a dense and sufficient task context.

Concept map and starter sequence

  1. Maintained knowledgedocumented concepts, rules, evidence and decisions
  2. Selection rulesdetermine relevance, representation and timing
  3. Task-specific contextsupports the current model call or agent action
  1. Current task statetask, requirements, data access, results and feedback
  2. Working Contextassembles selected knowledge, instructions, tools and state
  3. Task-specific contextenters the bounded context window as needed

Read in two paths: maintained knowledge flows through selection rules; current task state flows through the Working Context. Both meet in the task-specific context provided for the current action.

Separate prompt, context and knowledge base

The Project Knowledge Base preserves available project knowledge. The Working Context assembles what a concrete task requires. The prompt specifies the current action and expected result within that information state.

Treat the context window as bounded capacity

Longer context can suffer from position effects, distractors, conflicting material and obsolete state. Context rot names this observed degradation. The practical target is a dense, relevant and sufficient context.

Make relevant knowledge explicit

Knowledge acquisition draws from documents, data, established artefacts, interviews, workshops and observed workflows. It makes rules, exceptions, rationales and tacit decisions available for inspection and revision.

Maintain bounded Knowledge Documents

A Knowledge Document represents one subject or function through explicit concepts, relations, rules, evidence, uncertainty and provenance. Markdown is one practical serialisation; schemas, databases, TEI and RDF remain suitable where their semantics are required.

Select, structure and distil for a purpose

Compression reduces the amount of active material. Distillation also preserves the distinctions, conditions and open questions required for a specific purpose. Over-distillation removes provenance, uncertainty or operational detail that later work needs.

Combine persistent instructions, Skills, retrieval and tools

Instruction files govern recurring work. Skills operationalise reusable procedures. Retrieval selects external passages, while tools and MCP resources can expose files, databases or services. Retrieval failure and stale instructions must remain observable.

Join data context with method context

A file becomes useful to the model through the representation supplied by a parser, script, tool or multimodal system. A sound Working Context connects selected data with the methods, schemas, examples, permissions and checks needed to act on it.

Learning routes

  • Harness tutorial · available

    Project instructions and Working Context

    Follow the Claude Code tutorial routes for persistent project instructions and selected Working Context.

  • Related Module 04 route · available

    Curated Skills and reusable procedures

    The Agentic Engineering hub maintains the curated Skills entries. This module links to that collection without repeating it.

  • Deep dive · prepared

    Research Data and Knowledge Work

    The source-grounded route is prepared for integration; no public page is linked yet.

  • Deep dive · prepared

    Knowledge and Project Management

    The source-grounded route is prepared for integration; no public page is linked yet.

  • Workshop Web Module · available

    From research question to browser application

    Connects documented research context with requirements, tests and observable browser views.

Teaching material

Selected primary videos

Five of the six primary Module 03 videos are shown here. Research Mission Control remains a related route whose primary classification is Module 04.

Open all six primary Module 03 videos in the Video Library · Open the ordered Video Playlist

Representative workshops

All seven registered workshops include Module 03. These three profiles show knowledge and context work in distinct institutional settings.

Browse all workshop profiles