CueSpecKeep the context you need. Choose how to explain the task. Take a practical template into your next session.
Compare input methods, understand tool boundaries, and combine spoken context with exact text.
What should I say aloud and what should I type when asking an AI tool to refactor code?
Copy a refactor brief with exact symbols, allowed files and behavior checks, then follow a fictional helper-extraction example.
How can I tell whether dictation helps my actual coding requests rather than just producing more words?
Compare typing and dictation with a task-matched worksheet that includes preparation, correction, review and a consistent ready-to-send finish line.
Do I need to switch an AI model, a conversation, a coding task or a terminal?
Choose a workflow by what needs to change: the model, the conversation, the coding agent, the workspace, or the terminal.
Should I use typing, system dictation, a text-entry app or a conversation companion on Mac?
Choose typing, macOS Dictation or a supported conversation companion by destination, review needs and processing stages.
Trace unfamiliar code and give an agent the relevant evidence for a focused investigation.
How do I ask an agent to understand an unfamiliar codebase before editing it?
Trace one path through an unfamiliar repository, map file evidence and unknowns, and write a bounded follow-up change request.
What should I give an agent when local tests pass but CI fails?
Compare same-revision local and CI conditions, distinguish fixture-path and cache hypotheses, and request one discriminating check.
How much context should I give a coding agent to investigate one failure?
Select debugging evidence by the question it answers, with three context packets, item-level decisions, and a complete parser investigation request.
How do I turn one failing test into a repair request without blessing a wrong assertion?
Connect a failing assertion to its requirement, distinguish implementation and expectation errors, and bound a small repair with meaningful cases.
How do I brief a dependency upgrade while preserving caller compatibility?
Map direct and transitive dependency changes, preserve the caller contract, and specify compatibility checks and rollback conditions.
Turn interface feedback and data requirements into behavior you can check.
How can I turn screenshot feedback into a specific UI change and acceptance check?
Pair screenshot regions with spoken feedback, a copyable request template and three complete fictional UI change examples.
A product builder consolidates several people's observations and suggestions before authorizing one implementation request.
Keep three feedback sources separate, record decisions and one unresolved conflict, then build an agent brief from settled constraints.
A non-specialist describes a multi-event UI behavior: pending work, another user choice, cancellation and a late response.
Specify a six-event UI timeline, result ownership and cancellation behavior before changing how a panel handles late responses.
How do I separate a needed feature from the implementation I am suggesting?
Separate an export feature's user outcome from technical guesses, then compare two implementation plans against the same acceptance examples.
How do I ask an agent to repair form labels, keyboard operation and error recovery?
Repair a fictional form with complete before/after HTML, label and error mappings, a copyable agent request and a keyboard walkthrough.
When does Retry repeat the same request, and when does it start a new job?
A worked request-identity brief covering lost responses, deliberate second jobs, delayed replays and mismatched inputs.
What should Undo restore while preserving later changes?
Define Undo for a named archive action with a complete history, preserved later edits, conflict and failure cases, and a copyable agent brief.
What data and failure rules should a CSV import request specify?
Define CSV fields, encoding, duplicates and failure policy with nine original fixtures, actual local results and a copyable agent brief.
How do I describe an API change so old and new consumers can be checked?
Turn an API change into a checkable contract with request and response examples, compatibility boundaries and an OpenAPI 3.1.1 teaching document.
Who may read or change each record, and what happens when access changes?
Define record-level read and change permissions with a role matrix, adverse cases, membership withdrawal history and a bounded agent brief.
Resume a task, define dependencies, and bring separate changes together with clear ownership.
What should I preserve so I can return to an AI coding task after an interruption?
Keep a compact handoff card with the task, artifacts, decisions and next action, plus two filled examples for resuming AI coding work.
How do I split parallel agent work into independently reviewable deliverables?
Split a feature into four deliverables with a dependency graph, ownership cards, shared fixtures and a corrected parallel-work sequence.
How do I integrate parallel agent changes that touch shared files?
Combine two agent change proposals using a shared baseline, ownership table, configuration decisions, exact patches and combined checks.
What is isolated by a Git worktree and what still depends on conversation context?
Check code and conversation mappings separately with a two-task diagram, four mismatch cases, inspection commands and a filled example.
Match a change to its request and evidence, then explain it in a review, pull request, or developer guide.
How do I verify an AI code change against the original request?
Match a requested behavior to a fictional Python patch, example assertions and a review worksheet before accepting an AI code change.
How should I explain partial batch outcomes before deciding what to repeat?
Use a five-item outcome ledger to separate success, skipped, failed and unknown effects, then define a bounded correction and reconciliation plan.