Domain 1 · lesson
Prompting and Task Execution
Learning goal: translate a work need into instructions that are executable and testable. Core references emphasize clear instructions, examples, success criteria, and improvement driven by observed results. S07 S08 S09
Objective 1.1
Create effective prompts for business and technical tasks
Core content
A useful prompt identifies the desired outcome, audience, available information, constraints, and output format. A role can shape perspective or tone, but it cannot grant access to data or make unsupported claims true. Separate instructions from reference material. Headings and delimiters reduce ambiguity; XML is a tool, not a ritual. Examples are particularly useful when categories, tone, or edge cases are hard to describe. S08
Applied example
Instead of "summarize these bugs," ask for release impact, evidence for each finding, known owner or unassigned, and a table that separates verified facts from proposals. Explicitly prohibit invented dates. The result is now testable for coverage and accuracy rather than merely judged by how professional it sounds.
Exam decision rule
If acceptance criteria or audience are missing, make them explicit. If required data does not exist, request it or mark the absence; do not compensate with a more authoritative persona. Check: another reviewer should be able to explain how they will decide whether the output passes.
Objective 1.2
Break down complex requests
Core content
Split work when one stage depends on a verifiable result from an earlier stage, such as extracting facts, validating them, analyzing them, then drafting. Define the input, output, and review point for each stage. Decomposition improves fault localization, but every stage adds work and another opportunity to propagate an error. Not every task needs a chain. S08 S24
Applied example
To turn twenty interviews into a product proposal: extract problems with supporting quotes and anonymized speaker IDs; group themes without deleting exceptions; prioritize with agreed criteria; then draft a recommendation that preserves uncertainty. If extraction invents a complaint, the later recommendation can be perfectly coherent and still be false.
Exam decision rule
Choose the sequence that respects dependencies and preserves evidence. Independent source collection may run in parallel; synthesis must wait for the required inputs. Review should not appear only after a sensitive result has already been sent.
Objective 1.3
Iterate prompts to improve results
Core content
Start with explicit criteria and a set of cases. Preserve the initial version, observe concrete failures, change a relevant variable, and retest both the target failure and regression cases. Actionable feedback identifies a deviation: missing section, unsupported claim, wrong audience, incorrect category. "Make it better" is not a diagnosis. S07 S08 S09
Applied example
A ticket classifier confuses a price inquiry with a billing complaint. Add the distinction, a boundary example, and an insufficient information outcome. Test new cases from those classes and cases that were already correct. Do not test only the one ticket used to create the fix.
Exam decision rule
A local improvement is not proof of general improvement. If the root problem is an unreadable file, stale source, or missing permission, prompt iteration is not the causal remedy.
Objective 1.4
Adapt prompting strategy to the task type
Core content
Analysis needs criteria, units, assumptions, and evidence. Research needs sources, dates, scope, and contradiction handling. Writing needs audience, purpose, structure, and tone. Ideation benefits from diversity before selection, with constraints clearly separated from verified facts. S08 S16
Applied example
"Suggest names for this tool" supports creative exploration. "Which names are registered trademarks?" requires current research and validation. "Compare these two attached policies" requires reading both versions rather than giving a generic explanation of company policies.
Exam decision rule
Determine whether the gap is reasoning, information, or editorial direction. More thinking does not update stale evidence. Extensive research is unnecessary when the task is simply to rewrite user-provided text.