The Prompt Library
These prompts are starting points. The Chief of Staff should adapt them to the project’s actual repositories, tools, risks, and decision rights. “The Playbook Has to Fit in Human Memory” gives the situational index into these prompts, in “Setting It Up Takes an Afternoon and Five Commands.” Each prompt below is ready to paste.
Project state audit
Reconstruct current project state from specifications, decisions, issues, recent code changes, tests, acceptance evidence, meeting records, and commitments. Identify material contradictions, missing owners, stale assumptions, unsupported work, overdue checkpoints, and decisions requiring human judgment. Return only findings that could change what someone does next.
Question-to-requirement audit
Search recent conversations and meeting records for questions, suggestions, exploratory ideas, or customer requests that later appear as requirements, roadmap commitments, issues, or implementation work. Show the semantic transition and whether an explicit human decision authorized it.
Scope drift audit
Compare current implementation and active work with the latest approved specification. Identify additions, omissions, changed assumptions, or acceptance criteria that do not match. Distinguish implementation drift from likely stale specification. Escalate rather than silently correcting intent.
Customer chase audit
List customer or stakeholder commitments whose promised update or checkpoint is approaching or overdue. Identify whether the team has already prepared the evidence needed for a proactive update. Draft concise updates for human review.
Product evidence audit
Review current product claims and roadmap assumptions. For each consequential claim, identify supporting evidence, contradictory evidence, recency, source, and what observation would most weaken the claim. Distinguish repeated anecdote from broader evidence.
Experiment audit
List active experiments. For each, state the hypothesis, evidence being collected, stopping condition, current result, and decision the result is supposed to change. Flag experiments that cannot change a decision or that have continued after their stopping condition.
Weekly review
Compare last week’s approved outcomes and learning goals with actual results. Show completed and missed commitments, evidence learned, acceptance status, drift, blocked work, unresolved decisions, and proposed next-week outcomes. Recommend work to stop.
Meeting preparation
Prepare this meeting by showing what changed since the previous checkpoint, prior commitments, completed and incomplete actions, current evidence, changed assumptions, unresolved questions, scope changes, and the decisions humans need to make. Do not include routine status that participants can read themselves.
Meeting classification
Classify the conversation into decisions, questions, observations, hypotheses, ideas not approved, action items, owners, acceptance criteria, specification changes, and next checkpoint. Flag ambiguous statements and any place where a weaker category appears to be treated as a stronger one.
Work authorization gate
Before beginning substantial work, trace the requested work to current specification, explicit decision, approved experiment, defect, regulatory requirement, or operational necessity. If no valid trace exists, do not invent one. Explain the smallest clarification required.
Acceptance audit
Compare delivered behavior with current specification and acceptance criteria. For each criterion, show the evidence. Identify unspecified behavior, missing evidence, unresolved exceptions, and any human acceptance still required.
Management garbage collection
Inventory recurring reports, meetings, dashboards, forms, approvals, and manually maintained artifacts. For each, identify the failure it prevents, evidence it is used, duplicate sources, human maintenance cost, and a proposed stop condition. Rank candidates for removal.
Portfolio compression
Read subordinate project and product reviews. Identify only cross-project implications: conflicts, shared dependencies, duplicated work, repeated failures, common customer signals, resource contention, reusable capabilities, experiments to stop, and decisions requiring executive or capital-allocation attention.