The index/PromptThief Lab - PT-LAB-001

Legacy to agent-ready: Clarify before you act

One opt-in prototype. Existing prompt records and every other detail page remain unchanged.

Legacy - sequential chat

Ask one question at a time

Ask me one question at a time until you have enough information to succeed at the task.

Risk: useful in a chat loop, but it defines neither a planning budget nor what the agent should do after questioning.
Agent-ready - bounded brief

Brief the work, then clarify

TASK
[Describe the outcome you want.]

CONTEXT
[Attach or paste the facts, examples, and current source material the work depends on.]

TOOLS
[Name any tools the agent should use, or say none are required.]

CLARIFICATION BUDGET
Ask me up to 20 clarifying questions until you have enough information to succeed at the task. Group related questions when that reduces back-and-forth. Stop early when the remaining answers would not change the work.

BOUNDARIES
[Time, cost, sources, audience, privacy, and actions that need approval.]

DONE MEANS
[Define the deliverable and the evidence that proves it is complete.]

Dependencies

  • #The task and desired outcome
  • #Current source material or a clear statement that none is needed
  • #Available tools and permissions
  • #Hard boundaries and approval points

Done means

The agent asks only questions that can change the plan, stays within the ceiling, stops early when it has enough, and produces a brief with explicit inputs, tools, boundaries, deliverable, and acceptance evidence.

Test receipt

Status
Hypothesis queued for benchmark
Source
Christopher S. Penn, Sep 20, 2026
Compared
Legacy sequential instruction vs bounded package
Pass criteria
Better completeness and lower human cleanup without runaway questioning
Evidence
Prototype only. No performance claim yet.

Package manifest - v0.1

Version era: agent-ready · Task type: planning · Context: required · Tools: explicit · Clarification ceiling: 20 · Acceptance test: explicit · Model/harness: not yet benchmarked · Freshness: Sep 20, 2026 · Lint flags on legacy: unbounded completion, missing context, missing tools, missing definition of done.