Plugin Feedback

Plugin Feedback

Capture an idea, rough edge, or bug about a lab skill and file it as a GitHub issue on MusserLab/lab-claude-skills, so skill improvements reach the whole lab. Works whether or not the user has the gh CLI authenticated.

When this runs

• The user typed /plugin-feedback, or asked to report / suggest something about a skill or the plugin.
• Claude offered to capture skill friction (per the general-reminders standing instruction) and the user accepted.

You usually already have the context from the conversation — draft the report from it rather than interrogating the user.

Step 1 — Draft the report (don't interrogate)

Pull together, mostly from the conversation you already have:

• Which skill was involved (or "new skill idea" / "general" if none).
• Type: wrong behavior · missing case · activation problem · new-skill idea.
• What happened — the concrete rough edge, in one or two sentences.
• What was expected instead.
• Context (optional) — the relevant snippet of the conversation.

Keep it short and specific. Ask the user only for what you genuinely can't infer.

Step 2 — Confirm before filing (always)

Show the drafted title + body and ask the user to confirm or edit. Never file without asking.

Title format:

[skill-feedback] <skill-name>: <one-line summary>

Body template:

**Skill:** <name, or "new skill idea" / "general">
**Type:** wrong behavior | missing case | activation problem | new-skill idea

**What happened:**
<one or two sentences>

**What I expected:**
<one or two sentences>

**Context (optional):**
<relevant conversation snippet>

_Filed via the `plugin-feedback` skill._

Step 3 — File it

Write the confirmed body to a temporary file first (your scratchpad dir if you have one, else mktemp). Call its path BODY_FILE below.

Check for gh first:

gh auth status

If gh is authenticated — create the issue:

gh issue create \
  --repo MusserLab/lab-claude-skills \
  --title "[skill-feedback] <skill>: <summary>" \
  --body-file BODY_FILE \
  --label skill-feedback

If the command errors because the skill-feedback label doesn't exist in the repo, retry the same command without --label skill-feedback. Report the issue URL that gh prints back to the user.

If gh is missing or not authenticated — build a pre-filled "new issue" URL and give it to the user to click (no auth needed). URL-encode the title and body:

python3 - "$BODY_FILE" <<'PY'
import sys, urllib.parse
title = "[skill-feedback] <skill>: <summary>"
body  = open(sys.argv[1]).read()
base  = "https://github.com/MusserLab/lab-claude-skills/issues/new"
q = urllib.parse.urlencode({"title": title, "body": body, "labels": "skill-feedback"})
print(f"{base}?{q}")
PY

Give the user the resulting link: "Open this to file the issue (pre-filled): <URL>". If the body is very long (URL over ~7000 chars), trim the context snippet — GitHub rejects over-long URLs.

Common mistakes

• Auto-filing. Always confirm the drafted issue with the user first.
• Interrogating. You already have the conversation — draft from it; ask only for gaps.
• Assuming gh works. Check gh auth status; fall back to the pre-filled URL for users without gh.
• Over-filing. One issue per distinct idea. If the user raises several, ask which to file (or file separate issues).
• Wrong repo. Always MusserLab/lab-claude-skills, never the user's own project repo.