Close the frontier between what is decided and what is still open: the facts are gathered from the repo, the decisions go to the user in short rounds, each with a recommended answer. Use when a request is underspecified, the done-when is not stateable, or a rules/risk.md trigger is unresolved before code starts. Auto-triggers on: "grill me", "what have we decided", "ask me the hard questions".
Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.
What the file says
# Grill
A frontier is the line between what the repo already settles and what is still a
choice. Two failure modes sit on it, one at each end: asking a person for something
the code can answer, and building on a decision nobody noticed was a decision. This
walks the line in rounds and stops the moment the done-when is stateable.
`reference.md` holds the test that separates a fact from a decision, and the question
format with worked examples. Read it when you reach step 3.
## 1. State the frontier
Two short columns, filled from what you read and not from what you assume:
| Decided | Undecided |
| --- | --- |
| <what, with where it is decided> | <what, with who has to choose> |
Every left-hand cell cites: a `file:line`, a test, a committed record. Anything with
no citation goes right, including what looks obvious. A frontier nobody can cite is a
mood, not a frontier, and the right-hand column is the work.
## 2. Facts are gathered, never asked
Before a single question reaches the user:
- `devil digest` for the toolchain, the codemap, the untested surface.
- `devil facts` for the build, test and lint commands that actually run here.
…
The line at the top of this page pins the label git:20260930.4cc15b2, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.
GET https://markdownregistry.com/api/v1/artifacts/art_ltmf4mrcwkvwvtqy
GET https://markdownregistry.com/api/v1/resolve?ref=univers42/claude-deal-with-the-devil/grill
GET https://markdownregistry.com/api/v1/blob/79960415cdc7d898944bc954f31149440667f5351b045b9ccae7044cb4a0c78d
Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.
univers42/claude-deal-with-the-devil · skills/api-convention/SKILL.md · REST API conventions: resource-oriented and versioned routes, auth and access control on every surface, and one error…
univers42/claude-deal-with-the-devil · skills/api-endpoint/SKILL.md · Scaffold a new REST endpoint that mirrors the ones already there. Use when a route is missing, a handler needs wiring…
univers42/claude-deal-with-the-devil · skills/brainstorm/SKILL.md · Diverge wide on the options, then converge on evidence and cost, ending in one recommendation, a smallest experiment…
univers42/claude-deal-with-the-devil · skills/browser-testing/SKILL.md · Verifies a change in a real browser and returns the artifact: the steps, the viewport, the console. Use when a page is…
univers42/claude-deal-with-the-devil · skills/caveat/SKILL.md · Give every heuristic, sample, regex parser, cache, timeout and derived number the one line saying what it gets wrong…
univers42/claude-deal-with-the-devil · skills/commit-craft/SKILL.md · Turn a working tree into a history someone can read, revert and bisect. Use when changes are uncommitted, one commit…
univers42/claude-deal-with-the-devil · skills/context-budget/SKILL.md · Measure and cut what a config costs in context every session: always-on rules, invocable descriptions, skills that…
univers42/claude-deal-with-the-devil · skills/debug/SKILL.md · Find the actual cause of a failure instead of guessing at fixes. Use when a test fails, a build breaks, a crash has no…
univers42/claude-deal-with-the-devil · skills/design-review/SKILL.md · Judges an interface by hierarchy, rhythm, type, consistency and the states nobody built, then names the smallest fix…
univers42/claude-deal-with-the-devil · skills/doc-sync/SKILL.md · Find the documentation a change just made false and fix exactly that, examples copied from a passing run. Use when a…
univers42/claude-deal-with-the-devil · skills/frontend/SKILL.md · Builds UI that holds up at every width, in both themes, with real data, and for someone on a keyboard. Use when a page…
sethgammon/citadel · skills/grill/SKILL.md · Relentless one-question-at-a-time interview that sharpens a vague plan or design into shared understanding before any…
kerberosclaw/kc_ai_skills · grill/SKILL.md · Use when the user says 先討論 / wants to align on a fuzzy idea before building, or before any non-trivial implementation…
kanevry/session-orchestrator · skills/grill/SKILL.md · Use when the user wants a plan, design, or PRD stress-tested before any build — relentlessly interrogates one decision…
kanevry/session-orchestrator · .cursor/skills/grill/SKILL.md · Use when the user wants a plan, design, or PRD stress-tested before any build — relentlessly interrogates one decision…
kanevry/session-orchestrator · .codex-plugin/skills/grill/SKILL.md · Use when the user wants a plan, design, or PRD stress-tested before any build — relentlessly interrogates one decision…
kanevry/session-orchestrator · .agents/skills/grill/SKILL.md · Use when the user wants a plan, design, or PRD stress-tested before any build — relentlessly interrogates one decision…
technicalpickles/pickled-claude-plugins · plugins/stay-principled/skills/grill/SKILL.md · Walk a depth-first decision tree anchored to a project's documented principles before brainstorming opens options…
spair/task-pipeline · skills/grill/SKILL.md · Interrogate a plan or decision one question at a time before capture, keeping a decision-plus-rationale ledger, then…
