KB Grounded Answers: Stop Support Bots Inventing Prices
Answer customers only from your knowledge base, check every price and number in a reply before it goes out, and measure chatbot search quality per language (EN/RU/AR).
Free
Works with the AI Agents you already use
KB Grounded Answers: Stop Support Bots Inventing Prices
Example session with this skill installed
Customer asks: "Is irrigation expensive?" Draft reply: "Irrigation starts from $2,800 for a yard up to 5,000 sq ft, and this month we give 15% off if you order it with a lawn." Check it against our knowledge base before I send it.
- Read your context and instructions
- Compiled the kb grounded answers
grounded: false, unknown_numbers: ["15"]. The knowledge base has no 15% offer, so the agent removes it: "Irrigation starts from $2,800 for a yard up to 5,000 sq ft, with controller and rain sensor, installed in 2 to 4 days. Would you like a free site visit for an exact quote?" Re-check: grounded: true.
Connects securely to your tools. The creator never sees your data.
About this skill
Stop your support bot from inventing prices.
kb-grounded-answers makes your agent answer customer questions only from your company's knowledge base (a simple Markdown file), and checks every number in a draft reply against that base before it goes out: prices, terms, percentages, dates. An invented "15% off" gets caught. So does an English reply to a Russian question.
It also measures your chatbot's search quality per language (hit@1, hit@3, MRR). The bundled example shows keyword search scoring 100% on English questions and 0% on Russian and Arabic ones against an English base, and what to do about it.
What you get
- A CLI that needs only the Python standard library, with no API keys and no network calls. Commands: sections, search, check, eval, audit.
- An optional read-only MCP server with the same tools, for Claude Code, Cursor and Claude Desktop (works with mcp 1.x and 2.x).
- A clear answering workflow: search, then draft from the base only, then run the grounding check, then hand off to a human when needed.
- Eval and audit formats, an example knowledge base, 12 eval cases in 3 languages, and 12 tests.
Built from site-ai-consultant, a website AI consultant with RAG, evals, OpenTelemetry tracing and 78 tests: github.com/AbdAllAh950/site-ai-consultant
Good for: support and sales replies from a FAQ or price list, QA of a RAG chatbot, and multilingual customer service (EN/RU/AR).
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Search is keyword-based with simple stemming, so it does not match synonyms or questions in another language than the knowledge base (the skill tells the agent to translate the query). The grounding check covers numbers, not wording: a reply can still misstate a condition in words.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 17 days ago
- Free with an Agensi account
- Software
Python 3.9+
- Optional
- mcp Python package 1.9+
The mcp package is only for the MCP server.
Permissions
File Scopes
No network access, no environment variables, writes no files.