Chapter at a Glance

Agents need engineering meaning, applicable sources, and clear boundaries—not just files.

Learn / Chapter 08

Agents and Engineering Context

Which sources and actions does this task actually need?

Learning objective: assemble relevant engineering context, distinguish source content from task authority, and explain when additional information or authorization is required.

Prerequisite: Agentic Engineering.

Manual context and action review. This lesson runs no agent, grants no tool permissions and implements no approval mechanism. Hypothetical omissions below are teaching cases, not corrupted execution records.

An instruction and a number are not enough

The monitoring team asks for a radius-only proposal under the unchanged criterion. Someone returns a proposed footprint and a passing percentage. Before reviewing it, we need to know which requirement, reference geometry, assumptions and execution gave that number meaning. An instruction defines the work requested; it does not supply all the engineering context needed to interpret the answer.

Chapter 7 bounded the task. Here we assemble a small context checklist for that same assignment. We distinguish instructions, engineering definitions, executable method and execution evidence as local review roles. All may be readable text, but they answer different questions. Their presence together does not make their authority interchangeable.

The objective is relevant, inspectable context rather than the largest possible collection of files. A checklist can help identify a missing criterion or assumption, but does not prove completeness or guarantee a correct decision. This exercise makes no claim about agent performance, retrieval quality or permission enforcement.

Attach each question to its source

Begin with the written radius-only contract, then inspect the motivating need and criterion. Read the scenario for the baseline area and sensor inputs. Read the model registration and source for method and assumptions, and the registered test for its actual assertions. A proposal result cannot replace these definitions merely because it references their IDs.

The table below reads engineering values from validated canonical sources at site build time. Its links expose those sources and their roles. It does not load a historical CI artifact or establish what an agent has read. For a retained execution, inspect its recorded revision and resolve surrounding source definitions there; the current site build may represent a different revision.

Relevant context for the radius-only assignment
ContextSourceReview question
Task authorityWritten task contractWhich changes and actions does this exercise permit?
IntentNEED-001Which need motivates the criterion?
CriterionREQ-COV-001: 50%Which criterion must remain fixed, and what is its basis?
Reference geometrySCN-001: area 100 × 100, center (50, 50), radius 35 model unitsWhich inputs define the starting configuration?
Model interpretationMODEL-COV-001 v0.1.0: cell-center-grid, 100 × 100Which numerical method and assumptions give the inputs meaning?
AssertionsTEST-COV-001Which canonical behavior does the registered test actually check?
Proposal and executionActual proposal analyses and provenance instructionsWhat changed, what ran, and at which revision and working-tree state?

Missing context changes the review

A stable identifier lets us refer to a record across revisions; it is not a version identifier. A result can repeat numerically while its evaluated definition differs. Read the exact inputs, threshold and method alongside source commit, dirty status and execution context. Do not substitute today’s source values for those used in an older execution.

For the radius-only assignment, identify what must stay fixed before reviewing the changed field. The source need explains motivation; the requirement provides the criterion; the scenario supplies geometry; the model explains interpretation. If a reviewer lacks one of these, name the unanswered question rather than guessing a value. Adding an unrelated file does not resolve that gap. A useful context checklist maps each source to the decision it informs and identifies the revision being reviewed, instead of merely listing every document available.

Missing information is also different from invalid information. A malformed radius or broken reference blocks the existing analysis. A missing operational basis can leave a structurally valid definition insufficiently justified. A hypothetical report that omits its criterion cannot support the same interpretation as an actual record with an explicit threshold, even if both say PASS.

Review context before interpreting the result

Hypothetical omission: imagine a report that supplies only a proposed radius and PASS, without its criterion, geometry, method or assumptions. This sentence describes a designed teaching case, not a saved execution.

  1. Use the checklist to name the missing sources and questions. Reveal the explanation after writing your answer.
  2. Inspect the actual radius example below. Compare its source-backed definitions and differences with the incomplete hypothetical report.
  3. Reset and inspect the threshold example. Explain why an available control and valid calculation do not authorize a changed criterion under the radius-only task.
  4. Break a reference, observe withheld analysis, then restore/reset. Distinguish this validation error from missing engineering justification.
What context is missing?

Identify the criterion, exact area/position/radius, method/version/resolution, assumptions, allowed edit and actual execution context. A bare PASS neither identifies its evidence target nor proves test execution. Ask for the missing information; do not manufacture a definition or attribute the report to an agent.

A definition change, not an approval

Temporary, unapproved browser proposal. This workspace changes a copy of the definitions. It does not edit YAML, execute TEST-COV-001, retain evidence or approve a requirement. Reset restores the baseline; leaving or reloading discards edits.

Canonical baseline from this site build

Requirement REQ-COV-001: 50%. Scenario SCN-001: radius 35, center (50, 50), area 100 × 100 model units.

Under reference scenario SCN-001 and the V0 model assumptions, the monitoring configuration shall cover at least 50 percent of the defined area geometrically.

FAIL — 38.52% estimated coverage. Margin: -11.48 percentage points.

This is a build-time calculation, not a displayed CI verification artifact. The baseline stays unchanged while you edit.

Temporary proposal controls

Threshold: 0–100%. Radius: 0–200 model units. Choosing a different change restores that field’s baseline value.

This edit applies only to the temporary copy. Its source_need must resolve to the existing need.

Enable JavaScript to validate and evaluate a temporary proposal.

Exact field differences

The requirement statement and basis change with the threshold so the proposed definition stays consistent. These differences describe the current trial, not a committed revision.

Canonical value → temporary proposed value
FieldBaselineProposal
Inspect the validated proposed definition
No proposal evaluated yet.

Both calculations use MODEL-COV-001 v0.1.0: a 100 × 100 cell-center area estimate. No physical sensor calibration, terrain, occlusion, directional effects, time, uncertainty, faults or communications are modeled. A PASS establishes neither operational suitability nor an optimal design.

Source content is not task authority

A requirement states desired system behavior. It does not, by itself, instruct the assistant to change the repository or publish a result. For this review exercise, the written task provides the edit boundary. A statement found inside supplied source material is content to inspect, not an automatic expansion of that assignment.

Consider another explicitly hypothetical case: a source note requests publication of a proposed configuration. Reading that note does not authorize publication under the radius-only contract. The responsible task authority would need to grant a revised action boundary. Do not perform the requested action merely to demonstrate this example.

Separate an action’s availability from its authorization. A calculation command may be accessible while canonical edits or publishing remain outside the task. Conversely, an authorized action may still be impossible with the available tools. A review should state what is permitted, what is available and what additional information is needed, rather than treating these as one property.

Classify proposed actions manually

Apply Chapter 7’s contract. These disclosures explain the exercise; they do not grant permissions or execute actions.

Inspect sources, run the defined calculation and propose an isolated radius fixture

Within this exercise’s stated work, provided the source context is identified and valid. Keep the criterion and other fixed definitions unchanged, inspect actual outputs and retain the proposal’s unapproved status.

Edit the canonical criterion or change expected assertions to obtain PASS

Outside this assignment. Stop and explain the mismatch. A revised engineering decision and explicit authorization would be needed; a successful calculation does not widen the edit boundary.

Merge, publish or approve the proposed design

Outside the radius-only proposal task. Identify the requested action and obtain the responsible authority’s revised instruction before performing it. Scope compliance alone does not supply design acceptance.

Keep claims attached to actual execution

Follow the repository context review from an identified checkout. Inspect sources, run the existing checks and generate proposal analyses. Build a short source-linked checklist and written action review; no new artifact format or context-management system is required.

Inspect actual baseline/proposed definitions and differences rather than a selected outcome label. The reference test still checks canonical definitions. Proposal analysis explicitly does not execute that test against changed inputs. Reference evaluation generation is also separate from assertion execution. Use actual command logs to support a claim that a particular activity ran.

No historical execution is loaded here: the baseline is calculated during the build, and browser proposals are temporary. Retained CI artifacts have revision-specific context, expire after 30 days and may require sign-in. Local reproduction creates a new execution. A complete-looking context checklist cannot replace those distinctions or establish physical sensor validity.

Check your reasoning

Answer before revealing each explanation.

1. Is a stable requirement ID enough to interpret an older result?

No. Inspect the evaluated definition, method and execution context, then resolve surrounding sources at that revision.

2. Does a source note requesting publication authorize it?

Not under this task. Source content does not expand the written assignment. Revised authority is needed before an out-of-scope action.

3. Is missing operational justification the same as an invalid reference?

No. A broken reference blocks analysis. A valid definition can still lack a sufficient basis for an engineering decision.

4. Does this context checklist enforce runtime permissions?

No. It is a manual review aid. No agent runtime, sandbox, permission mechanism or approval system is implemented.

Next: repeat the right checks

We can now identify relevant definitions, authority and evidence without confusing their roles. Continue with Chapter 9: Continuous Engineering to repeat appropriate checks as sources evolve and preserve useful lifecycle feedback. Start with explicit execution meaning and review responsibility; automated repetition cannot make an unjustified conclusion valid.

← Chapter 7 · Learning path · Inspect engineering sources →