Supergrill — Documentation skill for Claude Code
Evidence-first interrogation protocol for Claude: recon, three depth levels, fact triage, resumable briefs.
How to install Supergrill
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Black-coffe/supergrill and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Supergrill does
Evidence-first interrogation protocol for Claude: recon, three depth levels, fact triage, resumable briefs. A markdown skill, no code.
Alternatives in Documentation
- 会话管理(Session Manager)需求文档(PRD / Markdown) — 目标:对 Codex / Claude Code 的本地会话记录进行可视化管理,并提供“一键复制 / 一键终端恢复”能力 31.7k ★
- Notion Research Documentation — Research Notion content and synthesize findings into structured briefs 14.6k ★
- Repo Recap — Generate a structured recap of the repository state: open PRs, open issues, recent releases, and executive sum 11.9k ★
README
Supergrill
**An evidence-first interrogation protocol for Claude.** Before you commit to a plan, an idea, a decision, an investment, or a text, Supergrill reads everything your environment already knows about it, then interviews you one question at a time until the weak points are on the table — with sources, contradictions, and a brief you can hand to the next stage.
It is a *skill*: a markdown protocol Claude follows. No code, no dependencies, no OS-specific tooling. It works in Claude Code, in IDE agents that read skills, and (pasted as instructions) in the Claude web and desktop apps.
/supergrill deep hard fund Series A in Acme
/supergrill quick should we rename the product
/supergrill super migrate billing to usage-based pricing
What it does differently
| Ordinary Q&A | Supergrill |
|---|---|
| Asks you what it could have read | Recon first. Files, prior briefs, memory, connected read-only sources, the web — before question one. |
| One size of conversation | Three depth levels with explicit budgets: quick (≤10 questions), deep (10–30, adaptive), super (staged, gated, multi-session). |
| Takes your numbers at face value | Fact triage on every answer. Load-bearing public facts are verified before the next question; the rest are graded at each wave. |
| Forgets what you said ten minutes ago | Contradiction pass every wave. Words vs words, words vs recon, words vs documents. Every hit becomes the next question. |
| Ends in chat scrollback | The brief is the session state. Written incrementally, resumable in a later session, exportable. |
| Praises the idea | Sycophancy is failure. "Great idea" is banned until the idea has survived. |
The three levels
quick |
deep |
super |
|
|---|---|---|---|
| Questions | up to 10 | 10–30 | 10–15 per stage, 3–7 stages |
| Time | 10–15 min | up to 30 min | several sessions |
| Recon | local only | + connected sources + web | full, with a reco |
Related Skills
Wiki Doctor
Diagnose and repair the active llmwiki installation and this repository's wiki — owned wiring first, then dete
Crewplane
Run reviewable, resumable coding-agent workflows across Claude Code, Codex, Gemini, GitHub Copilot CLI, or any
Auth.Md
An open protocol that lets agents register for services on behalf of users — discoverable through a Markdown f
SkillSafeWerkstatt
Portable evidence-linked Markdown wiki skills for Claude Code, Claude Cowork, and Codex
Competitor Event Map
An agent skill that maps which events, conferences and communities your competitors sponsor, at what commitmen
Dsh Mattpocock Skills Deck
Matt Pocock 技能面板(DSH Deck):安装即自带 25 个工程与效率技能(mattpocock/skills v1.2.3:triage、grilling、wayfinder、handoff 等),右侧面
Related Agents
Contributor Auditor
Contributor-trust triage specialist — depth-first review of an external PR for author signal, diff shape, repo
Web Recon
Read-only web reconnaissance — fetch and report on documentation pages, changelogs, release notes; URL fact-ch
Design Evaluator
Persistent evaluator on a design sprint team. Scores three designer proposals across three rounds against weig