Don't Anger Mario

Use this skill before opening an issue or contribution proposal in earendil-works/pi.

Rules

• Read CONTRIBUTING.md first.
• Use an official issue template; blank issues are disabled.
• Keep it short. If it does not fit on one screen, it is too long.
• Write in the user's own voice, not corporate/AI slop.
• Be concrete: what, why, and optionally how.
• Do not submit automatically.
• Before submitting, ask the user to confirm they carefully read the exact final draft and explicitly approve posting it.
• If proposing a change, use Contribution Proposal rather than Bug Report.
• If reporting a bug, include minimal repro and version.
• If opening a PR, only proceed after maintainer approval (lgtm for PRs; lgtmi is issues only).
• Do not send agent-only PRs. If the user cannot explain the change without the agent, stop.

Safe workflow

1. Inspect templates and contributing docs:

gh api repos/earendil-works/pi/contents/.github/ISSUE_TEMPLATE --jq '.[].name'
gh api repos/earendil-works/pi/contents/CONTRIBUTING.md --jq .content | base64 --decode

2. Draft the issue body in /tmp, never inline long Markdown through shell quoting:

cat > /tmp/pi-issue.md <<'EOF'
### What do you want to change?

...

### Why?

...

### How? (optional)

...
EOF

3. Show the exact final title and body to the user.
4. Ask for explicit approval with this wording:

Please read the draft carefully. Do you explicitly approve me posting this issue/proposal as written?

Do not proceed on vague approval like "ok" if the user has not seen the exact final draft.

5. Create only after explicit approval:

gh issue create \
  --repo earendil-works/pi \
  --template contribution.yml \
  --title "Short concrete title" \
  --body-file /tmp/pi-issue.md

Extra caution from prior tracker behavior

Known maintainer reactions that shape this workflow:

• Auto-close gate is intentional; maintainers review closed issues later.
• GitHub Actions outages can let the issue/PR gate fail and flood the repo.
• Maintainers dislike agent-only PRs because they take time to review and are often broken.
• "Clanker misunderstood the code" is a real failure mode; verify the claim before drafting.
• A label/comment like possibly-openclaw-clanker is a warning sign: shorten, verify, and make the human reasoning obvious.

Useful receipts:

• https://github.com/earendil-works/pi/issues/3589 — "very tired of clankers"
• https://github.com/earendil-works/pi/issues/5259 — "your clanker misunderstood"
• https://github.com/earendil-works/pi/pull/2666 — "You and your clanker are banned"
• https://github.com/earendil-works/pi/pull/5296 — "please do not send more agent only PRs"

Good contribution proposal shape

### What do you want to change?

One concrete change.

### Why?

The user-visible problem and why it matters.

### How? (optional)

Small implementation sketch, if useful. Avoid over-design.

Tone

Good:

Prompt templates support $1 and $ARGUMENTS, but optional arguments cannot have defaults. I want /next and /next 3 to expand cleanly without asking the model to interpret an empty argument.

Bad:

Implement advanced Bash-compatible templating with all parameter expansion semantics, conditionals, filters, and expression evaluation.