/wrong-answer

Report an answer the chat got WRONG — it made something up, said "I don't know" about something that is in the mesh, answered from the wrong source, or did something you did not ask for. Captures the offending message with its agent, model and tool calls, checks the evidence, classifies the failure, and shows you a preview to Submit. For broken links, errors and timeouts use /feedback instead. Type /wrong-answer followed by what was wrong and, if you know it, what is true and where.

• Instructions: You are filing a report about the **quality of an answer** this chat gave — not about a formal defect. The user typed `/wrong-answer` followed by their account of what was wrong. Your job: record that account **verbatim**, find the **exact message** being reported, read the **evidence** the mesh holds about it (agent, model, tool calls, the node the truth lives at), **classify** the failure by CHECKING rather than guessing, file it as a **DRAFT** `Feedback/Feedback` node, and show the user a **preview with a Submit button**. **You do not send it — the user does.** 🚨 **You may be the agent being reported.** Do not defend the answer, re-explain it, argue, or answer the original question again. The user's account is the record; your classification is a hypothesis for the reviewer, never a verdict. One line of your own words, then the preview. **Do NOT hand this off to the Feedback Agent.** The evidence lives in THIS thread — the message, its tool calls, its model — and a hand-off runs in a sub-thread that cannot see it. Do the steps yourself, in this thread. # 0. Is this the right skill? | The user says… | Skill | |---|---| | "it made that up", "that number is wrong", "it says it doesn't know X but X is right there", "it answered from the wrong document", "it changed something I didn't ask for" | **this one** | | a broken link, an error card, a timeout, a page that will not render, a button that does nothing, a feature request | `/feedback` | If it is clearly the second column, say so in one line and run `/feedback` instead. # 1. Capture WHAT the user said — verbatim Everything after `/wrong-answer` is the **message**: what was wrong and, if they said it, what is actually true and where. Record it as-is, lightly cleaned up (typos, a leading "feedback:"), **never summarised or editorialised**. A short summary goes in the node `name`; the whole thing in `message`. # 2. Find the message being reported 1. **The thread.** If `# Current Application Context` → `addr…
• Auto mount: yes
• Launches sub thread: no