clarify

Runs a four-stage clarification funnel with critic review before surfacing user decisions.

Updated May 31, 2026
One-click install
npx skills add https://github.com/AlexanderNarbaev/agi --skill clarify-alexandernarbaev
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: clarify
Source: https://github.com/AlexanderNarbaev/agi/tree/main/.opencode/skills/clarify
Command: npx skills add https://github.com/AlexanderNarbaev/agi --skill clarify-alexandernarbaev

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Ambiguous user requests lead to plans built on unstated assumptions or endless clarifying questions. This Skill structures how an architect agent inventories uncertainties, resolves what it can, and escalates only genuine user decisions. ## Core Features & Use Cases - Four-Stage Clarification Funnel: Inventories all material uncertainties, classifies each as self_resolved, critic_resolved, research_needed, user_decision, or deferred_nonblocking, consults a critic sounding board, then surfaces a structured decision packet. - Critic Verdict Mapping: Maps SoundingBoardVerdict outcomes (UNNECESSARY, RESOLVE, REPHRASE, APPROVED) to funnel actions with hard protections against dropping always-surface categories like scope, data loss, security, and breaking changes. - Assumption Recording and Timeout Handling: Records every resolved uncertainty as an explicit assumption and reclassifies stalled research items to user_decision after a configurable timeout. - Use Case: When asked to plan a feature with unclear migration and rollout implications, the architect runs this funnel to resolve defaults from context, consult the critic, and present only the decisions that genuinely require the user. ## Quick Start Ask the architect agent to plan an ambiguous feature request and it will load this protocol to run the clarification funnel before presenting any questions.

Frequently Asked Questions about clarify

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I handle ambiguous user requests in an AI planning agent?

Run a structured clarification funnel: inventory all material uncertainties, classify each as self-resolvable or user-facing, consult a critic before asking anything, then surface only genuine user decisions as a structured packet with recommended defaults.

How to reduce unnecessary clarifying questions from an AI agent?

Classify uncertainties and resolve them from the user request, spec, codebase, or recorded context before asking. A critic sounding board reviews candidate questions and drops, resolves, or rephrases them so only approved items reach the user.

What questions should always be surfaced to the user?

Always surface scope boundaries, data loss or destructive behavior, security and privacy risk, backward compatibility, breaking API changes, new dependencies, deprecations, cross-platform impact, cost tradeoffs, UX choices, rollout strategy, and QA gate strictness.

What happens when research for a clarification question times out?

If research does not complete within the configured research_needed_timeout_ms (default 300000ms), the item is automatically reclassified as user_decision with a note that research was incomplete, preventing the funnel from stalling.

Can the critic drop questions about destructive behavior?

No. Items in always-surface categories cannot receive UNNECESSARY or DROP verdicts; they are overridden to APPROVED. A runtime validation in the decision-packet assembly code is recommended to enforce this mechanically.