grill-with-docs
Skill0 starsChanged 6 months agoMITView on GitHub
Grilling session that challenges your plan against the existing domain model in this repo, sharpens terminology, and updates documentation (CONTEXT.md, ADRs) inline as decisions crystallise. Use when the user wants to stress-test a plan against the repo's locked vocabulary and architectural decisions.
195 other projects use a version of this, 53 of them edited.
Adopted by
Most recently taken up by
- Robo-Co-op/robobuilder-standardedited, first seen Nov 16, 2026
- devdeepjs/kobayashi-marutheir own version, first seen Nov 16, 2026
- lhotwll217/owner-operatortheir own version, first seen Nov 16, 2026
- philoserf/claude-code-configtheir own version, first seen Nov 15, 2026
- bmtrnavsky/disciplined-vibe-codingtheir own version, first seen Nov 15, 2026
- niki914/zafirotheir own version, first seen Nov 15, 2026
Used with
Projects with it also use setup-matt-pocock-skills, decision-mapping and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
GoogleSkillsetup-matt-pocock-skills
Configure this repo for the engineering skills — set up its issue tracker, triage label vocabulary, and domain doc layout. Run once before first use of the other engineering skills.
Also in 185 projects that use this
FabienDehopreSkilldecision-mapping
Turn a loose idea into a sequenced map of investigation tickets, then drive them to resolution one at a time.
Also in 9 projects that use this
LesaniSkillwriting-beats
Shape an article as a journey of beats, choose-your-own-adventure style. The user picks a starting beat from the raw material, you write only that beat, then offer options for where to pivot next, beat by beat, until the article reaches a natural end. Use when the user has raw material and wants to assemble it as a narrative rather than an argument.
Also in 65 projects that use this
liufeng1976Skillsetup-ts-deep-modules
Wire dependency-cruiser into a TypeScript repo so each package is a deep module — implementation hidden in subfolders, reachable only through its entry-point files. User-invoked.
Also in 42 projects that use this
simo97Skillmigrate-to-shoehorn
Migrate test files from `as` type assertions to @total-typescript/shoehorn. Use when user mentions shoehorn, wants to replace `as` in tests, or needs partial test data.
Also in 67 projects that use this
LesaniSkillwriting-shape
Take a markdown file of raw material and shape it into an article through a conversational session — drafting candidate openings, growing the piece paragraph by paragraph, arguing about format (lists, tables, callouts, quotes) at each step. Use when the user has a pile of notes, fragments, or a rough draft and wants help turning it into something publishable.
Also in 63 projects that use this
- +1 more from Lesani
Other versions195 projects use a version of this.
Word for word · 0
None found.
Edited · 53
- bholmesdev/hubble.md★ 1.5k32% same
- sanity-io/next-sanity★ 95632% same
- victorgarciaesgi/regle★ 50832% same
- amazingloft999-droid/mattpocock-skills★ 42234% same
- openvetta/open-vetta★ 29932% same
- InditexTech/weavejs★ 26632% same
- brickbots/PiFinder★ 25632% same
- yamcodes/arkenv★ 15132% same
- alex-on-ai/WebReaper★ 14932% same
- calf-ai/calfkit-sdk★ 14632% same
43 more not listed
Wrote their own version · 142
- mattpocock/skills★ 325ksame name
- QuivrHQ/quivr★ 40ksame name
- The-Vibe-Company/quivr★ 40ksame name
- GreyDGL/PentestGPT★ 16ksame name
- formbricks/formbricks★ 13ksame name
- ArcReel/ArcReel★ 6.6ksame name
- vinvcn/mattpocock-skills-zh-CN★ 5.8ksame name
- withcoral/coral★ 4.9ksame name
- bestofjs/bestofjs★ 3.1ksame name
- CelestoAI/SmolVM★ 1.2ksame name
132 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository18 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- audit-docs-semanticSkill
- audit-docsSkill
- audit-harness-coverageSkill
- budget-statusSkill
- cavemanSkill
- configurationSkill
- critic-reviewSkill
- doc-patternSkill
- error-taxonomySkill
- grill-meSkill
- new-docSkill
- python-failure-modesSkill
- review-changesSkill
- review-docSkill
- rust-failure-modesSkill
- verification-before-completionSkill
- zoom-outSkill
The file
We split this skill into the rules it gives an agent
9 rules. 0 are common, 5 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 181 edited versions we read keep these.
5
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 19
Ask one question at a time, waiting for feedback before continuing.
- Distinctive: line 23
Before the first question, scan:
- Distinctive: line 31
Treat this scan as the prerequisite — the domain-awareness phase.
- Very specific: line 35
When the user uses a term that conflicts with the canonical vocabulary in the relevant CONTEXT.md / shared-kernel doc, call it out …
- Very specific: line 39
When the user uses vague or overloaded terms, propose a precise canonical term from the existing vocabulary first.
- Distinctive: line 45
When domain relationships are being discussed, stress-test with specific scenarios.
- Very specific: line 53
When a term is resolved, update the relevant CONTEXT.md right there.
- Very specific: line 69
If any of the three is missing, skip the ADR — vocabulary updates alone are sufficient.
- Distinctive: line 75
When the interview converges (no more open branches, all terms resolved, all decisions either documented in CONTEXT.md or escalated to …
“Common” is measured: kept by at least 6 in 10 of the 181 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/grill-with-docs
Requires the consuming repo to have at least one of: a root CONTEXT.md, per-subsystem CONTEXT.md files, or docs/decisions/ ADRs. The foundation cut ships no CONTEXT.md or ADRs — adopt the conventions in CONTEXT-FORMAT.md and ADR-FORMAT.md first.
Same interview shape as /grill-me, but every question stress-tests against the existing repo documentation: any CONTEXT-MAP.md at the root, per-subsystem CONTEXT.md files, shared-kernel docs, locked design decisions, and ADRs under docs/decisions/. Updates the relevant CONTEXT.md (vocabulary) and writes new ADRs (locked decisions) inline as decisions crystallise.
This skill is operator-fired only (writes to docs/ + meaningful LLM cost from the extended interview session). Every question + recommendation is an LLM-orchestrated turn.
When invoked: /grill-with-docs
Interview the user relentlessly about every aspect of the plan until you reach shared understanding. Walk down each branch of the design tree, resolving dependencies one-by-one. For each question, recommend an answer.
Ask one question at a time, waiting for feedback before continuing. If a question can be answered by exploring the codebase or reading the relevant CONTEXT.md / ADR, explore instead of asking.
1. Read the existing language first
Before the first question, scan:
- The root
CONTEXT-MAP.md(if present) — to identify which subsystems / bounded contexts the plan touches. - The relevant per-subsystem
CONTEXT.md(or equivalent) for each touched context — to absorb the canonical vocabulary and the_Avoid_:synonyms. - The shared-kernel doc (if present) — for primitive newtypes referenced across multiple contexts.
- The locked design-decisions doc — architectural commitments that constrain the plan.
docs/decisions/— dated ADR amendments.
Treat this scan as the prerequisite — the domain-awareness phase. Do not start interviewing until you can name the relevant contexts and at least the locked decisions that the plan touches.
2. Challenge against the glossary
When the user uses a term that conflicts with the canonical vocabulary in the relevant CONTEXT.md / shared-kernel doc, call it out immediately. Recommend the canonical term, cite where it lives, and ask the user to confirm before proceeding.
3. Sharpen fuzzy language
When the user uses vague or overloaded terms, propose a precise canonical term from the existing vocabulary first. Adding a new term should be a last resort — and if added, it must land in the relevant CONTEXT.md's Language section in the same session.
Watch for words that look domain-specific but mean different things in different subsystems (a "channel", "limit", "request", "intent" may each be defined by the local CONTEXT.md; cross-checking is what makes the grilling useful).
4. Discuss concrete scenarios
When domain relationships are being discussed, stress-test with specific scenarios. Invent edge cases that probe boundaries between concepts. The point is to surface the rules a clean abstract description hides.
5. Cross-reference with code
When the user states how something works, check whether the existing source / CONTEXT / ADR agrees. If you find a contradiction, surface it — quote the conflicting doc line, ask whether the plan is to override (which would need an ADR) or whether the user has misremembered the rule.
6. Update CONTEXT.md inline
When a term is resolved, update the relevant CONTEXT.md right there. Don't batch updates — capture them as they happen. Use the format in CONTEXT-FORMAT.md.
Discipline:
- Decide the owner context before adding the term. Composite types stay with their owner; only atomic newtypes referenced across all contexts are kernel candidates.
- Mirror the addition in the source's surface form: a new term in the subsystem's
CONTEXT.mdLanguage section + a matching_Avoid_:line for any synonym you ruled out during the session. - Operator review during
/grill-with-docsis the discipline that closes the loop: when a synonym you ruled out shows up in source later, catch it in the next interview session and rename it together./audit-docsdoes not enforce vocabulary discipline — only structural drift (D1–D11).
7. Offer ADRs sparingly
Offer to write an ADR only when all three are true:
- Hard to reverse — cost of changing your mind later is meaningful.
- Surprising without context — a future reader will wonder "why did they do it this way?"
- The result of a real trade-off — there were genuine alternatives and you picked one for specific reasons.
If any of the three is missing, skip the ADR — vocabulary updates alone are sufficient. When all three hold, write the ADR using the format in ADR-FORMAT.md, into docs/decisions/<YYYY-MM-DD>-NNN-<slug>.md (next dense NNN).
By convention this harness uses a single ADR home (docs/decisions/). Consuming repos may use docs/adr/ or per-subsystem ADR directories — adapt the refusal rule to your repo's convention. Most ADRs in a multi-context codebase end up being cross-cutting, which is why the single-home default works for many repos.
8. End-of-session summary
When the interview converges (no more open branches, all terms resolved, all decisions either documented in CONTEXT.md or escalated to ADRs), produce a one-paragraph summary:
- Contexts touched + vocabulary added (per file).
- ADRs written (with paths) or
none — no decision met all three ADR criteria. - Open follow-ups (if any) for the operator to decide later.
Operator escape valve
Drive steps 1-8 above turn-by-turn through chat (Edit/Write tools are not locked; only the slash is).
Failure modes
| Failure | Detection | Mitigation |
| Plan touches a context the interviewer hasn't read yet | The interviewer makes a recommendation that contradicts an existing CONTEXT.md term it didn't see | Operator catches it in review; interviewer re-reads the missed CONTEXT.md and re-frames the recommendation. Discipline: step 1 is non-negotiable — never skip the language scan. |
| Term resolved in session but CONTEXT.md never updated | Operator notices in a subsequent /grill-with-docs interview that source uses a synonym the CONTEXT.md should have ruled out | The interview itself is the gate — no automated check fires. Discipline: update CONTEXT.md during the session per step 6, not after. |
| Operator wants an ADR for a decision that fails the three-criteria gate | Interviewer offers, operator accepts; resulting ADR is noise that will get retired | Reject offers proactively — when in doubt, vocabulary update only; no ADR. |
| ADR home convention mismatch | Operator writes an ADR to a path that conflicts with the repo's configured ADR home (e.g., the repo uses docs/decisions/ and the operator writes to <subsystem>/docs/adr/) | Surface the mismatch and ask the operator which convention this repo follows; adapt to it. |
Forbidden
- Skipping step 1 (language scan) — interviewing without absorbing the existing vocabulary contradicts the skill's purpose.
- Adding a new term to CONTEXT.md without checking whether an existing term already covers it.
- Writing an ADR to a path that conflicts with the repo's ADR-home convention — confirm the convention with the operator before writing.
- Updating invariant docs (e.g.,
docs/harness/components/<x>/SPEC.md) with new vocabulary — those are invariant docs; vocabulary lives inCONTEXT.md; rules in the SPEC cite the vocabulary, they don't define it. - Auto-merging or auto-committing — this skill produces edits in the working tree; the operator decides when to stage / commit / PR.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

