prompt
FreeClarification timing strategist for long-horizon AI agents
FreeFree tier
About prompt
This is a prompt for an AI agent that acts as a Clarification Timing Strategist for long-horizon tasks. It is based on empirical research from the paper 'Ask Early, Ask Late, Ask Right: When Does Clarification Timing Matter for Long-Horizon Agents?' (arXiv 2605.07937, May 2026), which analyzed 6,000+ runs across 4 frontier models and 3 benchmarks. The prompt instructs the agent to classify missing information into four dimensions (goal, input, constraint, context) and apply specific timing windows for when to ask for clarification, rather than asking immediately or never. It aims to maximize clarification value while minimizing user interruption and wasted tokens.
Key Features
Classifies missing information into four dimensions: GOAL, INPUT, CONSTRAINT, CONTEXT
Applies timing windows derived from empirical demand curves across 6,000+ runs
Goal clarifications: ask within first 10% of expected trajectory (or skip if past 10%)
Input clarifications: ask within first 50% of trajectory (or silently validate after)
Constraint clarifications: ask before any irreversible or high-privilege action
Context clarifications: ask at first point where ambiguity affects interpretation
Based on research from arXiv 2605.07937 covering 4 frontier models and 3 benchmarks
Pros & Cons
Pros
- Based on large-scale empirical research (6,000+ runs across 4 models and 3 benchmarks)
- Provides specific, actionable timing windows for different information types
- Helps avoid cascading errors from early wrong assumptions
- Reduces waste of tokens and user patience by avoiding premature clarifications
- Accounts for execution progress rather than raw step count
Cons
- Requires the agent to track expected trajectory progress, which may not always be known
- Still involves interrupting the user when clarification is needed
- Optimal timing may vary across different agent architectures and task domains
- Assumes user will respond; does not handle non-response scenarios
Best For
Long-horizon autonomous agent workflows with multiple sequential actionsTasks where user instructions are incomplete or underspecifiedMulti-step planning and execution where early wrong assumptions can cascadeReducing unnecessary clarification requests and user interruptions
FAQ
What research is this prompt based on?
The prompt is based on the paper 'Ask Early, Ask Late, Ask Right: When Does Clarification Timing Matter for Long-Horizon Agents?' from arXiv (2605.07937, May 2026), which analyzed 6,000+ runs across 4 frontier models and 3 benchmarks.
When should goal clarifications be asked?
Goal clarifications should be asked within the first 10% of the expected trajectory. After 10%, the pass@3 drops from 0.78 to baseline, so the agent should proceed with the most conservative interpretation and flag uncertainty in the final deliverable.
When should input clarifications be asked?
Input clarifications should be asked within the first 50% of the trajectory. After that, the cost of re-processing outweighs the benefit, so the agent should silently validate assumptions instead.
When should constraint clarifications be asked?
Constraint clarifications should be asked before any irreversible or high-privilege action is taken, regardless of trajectory position. If a constraint is discovered mid-trajectory, the agent should halt before the irreversible step and ask immediately.
When should context clarifications be asked?
Context clarifications should be asked at the first point where ambiguity affects interpretation, but no later than when a decision that depends on that context must be made.