fangdididi

Answer Standing — AI skill for Claude Code

AI community

Agent skill: measure how AI assistants (ChatGPT, Claude, Gemini, DeepSeek, Doubao.

How to install Answer Standing

This entry records only its repository, not the path inside it, so there is no exact command to give. Open fangdididi/answer-standing and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Answer Standing does

Agent skill: measure how AI assistants (ChatGPT, Claude, Gemini, DeepSeek, Doubao...) describe your product with repeated sampling and confidence ranges, grade every claim, trace errors to sources, fix and re-test. EN + 中文

Alternatives in AI

AIHawk — Open-source AI browser agent: describe any task in plain language and it autonomously browses, clicks, types a 30.3k ★
Firm — Convene your standing AI staff — memos on every beat, a board session without you, minutes with dissent preser 1.3k ★
Describe Component — Write a six-section AI-optimised prose description of one component for Figma MCP and LLMs 194 ★

README

**English** | [简体中文](README.zh-CN.md)

Answer Standing

**Find out how AI assistants really describe your product, prove it with numbers, and fix wrong answers at their source.**

When someone asks ChatGPT, Claude, Gemini, Perplexity, Copilot, DeepSeek, Doubao or Kimi "what's the best tool for X?", the answer names a few products and states a few facts. Answer Standing is an agent skill that measures whether your product (or company, project, or your own name) is in those answers, whether what is said is true, where the wrong statements come from, and whether your fixes actually changed anything.

It treats the assistant as a witness: ask the same questions several times, record every answer verbatim, grade each statement against verified facts, trace errors to pages and listings, repair those sources, then re-measure with the same protocol and an honest statistical test.

Why it exists

One screenshot is not a measurement. Answers vary by run, mode, language and region. The skill samples repeatedly and reports rates with 95% ranges.
Being mentioned is not enough. A mention with the wrong price or a namesake's scandal can hurt more than silence. Every claim gets a verdict.
You can't edit an answer, only its sources. Every wrong claim ends in a concrete fix on a concrete URL and a date to re-check.
No wishful thinking. A built-in comparison tells you whether a change is real, maybe, or just noise.

Features

Truth sheet with as_of dates, public URLs, namesakes and retired facts
Four-family question deck (category, comparison, direct, problem) with a frozen core set
Fixed sampling protocol: fresh sessions, mode tracking, 3+ repeats, raw answers archived
Per-answer reading: mention, position, framing (recommended / neutral / warned), share of voice
Six claim verdicts: correct, stale, wrong, invented, conflated, unverifiable
Source tracing ladder from cited links to "model memory"
Source-first repair plan with honesty guar