grill-via-web

name

grill-via-web

description

Pop a clickable ChatGPT and Claude.ai URL that prefills a grill-me prompt referencing a public GitHub issue, so a human can finish the grill in a browser tab. Use when the user wants to grill an issue in a web LLM (no API, no local Claude needed). Trigger phrases include "grill via web", "grill issue N in chatgpt", "/grill-via-web", "give me a chatgpt url to grill <issue>", or "pop a grill URL for issue N".

       __                       ┌──────────────────────────┐
  ___ ( o)>    issue URL ──────▶│   /grill-via-web         │
  \  =._)                       │                          │
   `---'   URL-encode prompt ──▶│   pop two Chrome URLs    │
    呷呷~                       │   ChatGPT + Claude.ai    │
                                └─────────────┬────────────┘
                                              │
              human clicks Chrome ────────────┘
              LLM fetches mattpocock/skills/grill-me
              LLM grills 1 question per turn
              human answers, repeat until convergence

0. Pre-grill claim (cross-host mutex, mandatory)

Per docs/PRE-GRILL-CLAIM.md, before generating any Chrome URL the agent must:

make a comment claiming we have started grilling this issue and add tag "grilling"

Concretely (in the same order):

1. gh issue view <N> --json labels -q '.labels[].name' | grep -q '^grilling$' && exit — if grilling already present, another agent is on it; post 🚦 deferred: grill already in progress (grilling tag is set) and stop. Do NOT force-remove.
2. gh issue view <N> --json labels -q '.labels[].name' | grep -q '^grill-ready$' && exit — if the issue is already grilled, do not re-grill; route the user to /fixed-flow-driver instead.
3. gh issue comment <N> --body "🍳 grill picked up at <ISO timestamp> on <hostname> via /grill-via-web. Following docs/PRE-GRILL-CLAIM.md."
4. gh issue edit <N> --add-label grilling

Only after both the comment AND the label succeed: proceed to §1 URL generation.

If gh issue edit fails because the label grilling does not exist on the repo, post ⛔ grilling label missing on repo; ask a maintainer to create it via gh api repos/<owner>/<repo>/labels --method POST -f name=grilling and stop. Do NOT auto-create the label.

When the human pastes the grill output back into the issue (later, after the browser-side grilling finishes), the same human is responsible for gh issue edit <N> --remove-label grilling --add-label grill-ready to swap the lock to the next-phase label.

What it does

Takes a public GitHub issue URL (default scope: libz-renlab-ai/TeamBrain), URL-encodes a fixed prompt template, and outputs two clickable URLs:

follow instructions in https://github.com/mattpocock/skills/blob/main/skills/productivity/grill-me/SKILL.md  and grill me for this : <ISSUE_URL>

into:

• https://chatgpt.com/?q=<encoded>
• https://claude.ai/new?q=<encoded>

Only <ISSUE_URL> varies. The grill-me playbook URL is pinned to mattpocock's public skills repo so any anonymous LLM fetcher can read it.

Why URLs (not API / not local Claude)

Human-in-the-loop, by design. The skill does not run the grill — it only pops a Chrome URL. Human:

1. Clicks the URL → ChatGPT/Claude tab opens with prompt prefilled.
2. Hits send → LLM fetches the grill-me skill + the issue.
3. LLM asks one question → human answers → next question.
4. Done. Move to next issue. New URL. Repeat.

No auth, no API key, no rate limit, no agent loop. Cheap and throttle-free.

Template (do NOT change this string in code)

follow instructions in https://github.com/mattpocock/skills/blob/main/skills/productivity/grill-me/SKILL.md  and grill me for this : <ISSUE_URL>

Note the two spaces before and — preserved from upstream usage so URLs round-trip with anything that compares strings byte-for-byte.

Generator (bash one-liner)

ISSUE_URL="$1"
PROMPT="follow instructions in https://github.com/mattpocock/skills/blob/main/skills/productivity/grill-me/SKILL.md  and grill me for this : ${ISSUE_URL}"
ENC=$(printf '%s' "$PROMPT" | jq -sRr @uri | sed 's/%20/+/g')
echo "ChatGPT:  https://chatgpt.com/?q=${ENC}"
echo "Claude:   https://claude.ai/new?q=${ENC}"

If jq is not available, swap to python:

ENC=$(python3 -c "import sys, urllib.parse; print(urllib.parse.quote_plus(sys.argv[1]))" "$PROMPT")

When to invoke

User types or says any of:

• /grill-via-web
• grill via web
• grill me on issue N
• grill issue N in chatgpt
• give me a chatgpt url to grill issue N
• pop a grill URL for issue N

Output format

1. Two clickable markdown links (ChatGPT first, Claude second).
2. Three-line ASCII flow showing the human-in-loop step.
3. Done. No extra commentary.

Caveats

1. Public issues only. Anonymous web LLMs cannot read private repos. For private TeamBrain issues, fall back to inline-paste flow.
2. ChatGPT does not auto-submit. It pre-fills the input box; the human clicks send.
3. Claude.ai ?q= may be deprecated post-2025-10. If the input field stays empty, switch to ChatGPT or paste manually. See Claude Help Center.
4. URL length budget < 2000 chars. This skill does not embed the issue body — the LLM fetches it from GitHub via the URL reference. If the issue is private or the URL is gigantic, this skill won't work.

Example

Input:  https://github.com/libz-renlab-ai/TeamBrain/issues/276

Output:
  ChatGPT:  https://chatgpt.com/?q=follow+instructions+in+https%3A%2F%2Fgithub.com%2Fmattpocock%2Fskills%2Fblob%2Fmain%2Fskills%2Fproductivity%2Fgrill-me%2FSKILL.md++and+grill+me+for+this+%3A+https%3A%2F%2Fgithub.com%2Flibz-renlab-ai%2FTeamBrain%2Fissues%2F276
  Claude:   https://claude.ai/new?q=follow+instructions+in+https%3A%2F%2Fgithub.com%2Fmattpocock%2Fskills%2Fblob%2Fmain%2Fskills%2Fproductivity%2Fgrill-me%2FSKILL.md++and+grill+me+for+this+%3A+https%3A%2F%2Fgithub.com%2Flibz-renlab-ai%2FTeamBrain%2Fissues%2F276