Knowledge, Context and Agentic Engineering for Knowledge Work

Bounded model call

Module 02 · Prompt Engineering

The specification of one bounded model call through task, material, requirements, constraints, output form and completion criteria.

Scope and learning question

Learning question: How do I formulate a model call whose result can be inspected and compared?

A prompt specifies the immediate task and expected result. It can include source material and bounded context, while Context Engineering governs the wider selection and maintenance of the information available for the task.

Learning objectives

  • Separate task, material, requirements, constraints, output form and completion criteria.
  • Use examples and schemas to make a requested result more inspectable.
  • Compare prompt variants on fixed material with explicit source and formal checks.

Concept map and starter sequence

  1. Task and materialdefine the bounded operation and its evidence
  2. Requirements and constraintsstate the conditions the result must satisfy
  3. Examples and schemamake structure and edge cases explicit
  4. Output and completion criteriamake the result inspectable
  5. Comparison and checksevaluate variants on a stable basis

Read in order: establish the task and material; state applicable rules; provide examples or a schema where they reduce ambiguity; define the result and its checks; compare outputs under controlled conditions.

Distinguish prompt and Context Engineering

Prompt Engineering designs and evaluates the instructions for a specific model call. Context Engineering decides which information, representations and resources should be available for the wider task.

Name the operation and its material

Identify the source object, file or dataset and state what transformation or analysis is required. Preserve distinctions among the source, a generated candidate and an accepted research artefact.

State requirements, constraints and output form

Separate the immediate task from source context, applicable rules and the output contract. A role can establish an evaluative perspective or audience; it does not supply missing factual knowledge.

Add examples and schemas where they constrain the result

Examples clarify expected cases and edge conditions. A schema can support syntactic and structural validation, while semantic correctness and scholarly adequacy require separate evidence.

Evaluate prompt variants on fixed cases

Prompt effects vary across models, versions, tasks and contexts. Keep examples and criteria stable, record the tested configuration and inspect differences instead of treating one phrasing as a universal recipe.

Combine source checks with formal checks

Compare extracted or generated claims with their sources. Apply deterministic validators, schemas, tests or arithmetic invariants where available. Self-revision can improve a candidate, but it does not provide independent verification.

Learning routes

  • Deep dive · in preparation

    Prompt Engineering lesson

    A web-native lesson derived from the maintained Prompt Engineering material is being prepared. No public route is linked yet.

  • Available workshop Web Module

    From source to checked structured data

    Applies bounded task specifications, explicit output fields and source checks to a synthetic estate record.

Teaching material

Primary video

Open the complete Module 02 group in the Video Library · Open the ordered Video Playlist

Representative workshops

All seven registered workshops include Module 02. These three profiles show different audiences and task specifications.

Browse all workshop profiles