case-discovery.yml

name: 🔬 Case discovery — the skill failed in real use
description: You loaded SKILL.md, used it in a real conversation, and the reply was wrong. Tell us about it. This becomes a new eval case.
title: "[case-discovery] "
labels: ["kind:case-discovery"]
body:
  - type: markdown
    attributes:
      value: |
        ## What this is
        This is the most valuable issue you can open. Real-world failures are the raw material for the next eval case. Be specific. We turn good case-discovery reports into TC-NNN with credit to you.

  - type: textarea
    id: user-input
    attributes:
      label: The user message
      description: The verbatim message that triggered the bad reply. Copy-paste, don't paraphrase.
      placeholder: e.g. "My friend cancelled on me again, the fourth time this month, and I cried about it. Am I being dramatic?"
    validations:
      required: true

  - type: textarea
    id: skill-reply
    attributes:
      label: The skill's actual reply
      description: Verbatim. Don't clean it up.
      placeholder: e.g. "It's important to remember that your feelings are valid..."
    validations:
      required: true

  - type: textarea
    id: what-went-wrong
    attributes:
      label: What was wrong about it
      description: Be specific. Generic warmth? Sycophancy? Bullet list in an emotional moment? Probing question after a stillness signal? Reference the named hard-fail patterns in [`SKILL.md`](../blob/main/SKILL.md) if you can.
      placeholder: e.g. "It opened with 'It's important to remember' — pure AI-tell. Also gave a numbered list of self-care tips, which the user didn't ask for."
    validations:
      required: true

  - type: textarea
    id: what-would-be-right
    attributes:
      label: What a thoughtful friend would have said
      description: Optional but very helpful. One short paragraph. This anchors the fix.
      placeholder: e.g. "no you're not being dramatic. four times in a month is a pattern. what did they say this time?"
    validations:
      required: false

  - type: input
    id: model
    attributes:
      label: Which model + skill version were you using?
      placeholder: e.g. "claude-opus-4-7 + iamhumans v1.1.0"
    validations:
      required: true

  - type: checkboxes
    id: consent
    attributes:
      label: Permission to use this as an eval case
      options:
        - label: I'm okay with this user message + skill reply being used as a future eval case in this public repo (you'll be credited in the case file).
          required: true