agentleFS
drosamdrosam/agents/skills/grill-with-docs/SKILL.md

grill-with-docs

Skill1 starsChanged 2 days agoMITView on GitHub

Stress-tests a plan, design, or decision through a dependency-aware repository interview. Persists compaction-safe working notes and records settled domain terms and approved ADRs in the worktree-shared ~/.agent-work/<repo-key>/ work area. Use when the user says "grill me", "grill me with docs", wants assumptions challenged or a grooming session, or wants to document a repo's domain language.

195 other projects use a version of this.

Adopted by

Most recently taken up by

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.
  • GoogleSkill

    setup-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

  • FabienDehopreSkill

    decision-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

  • LesaniSkill

    writing-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

  • liufeng1976Skill

    setup-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

  • simo97Skill

    migrate-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

  • LesaniSkill

    writing-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

The file

We split this skill into the rules it gives an agent

11 rules. 0 are common, 8 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 181 edited versions we read keep these.

  • 8

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read references/notes-format.md.

    line 18
  2. Distinctive:

    Before the interview starts, capture all relevant information already present in the active session: objective, scope, constraints, …

    line 19
  3. Distinctive:

    If the notes write is denied or fails, stop before the interview and report the recovery blocker.

    line 20
  4. Distinctive:

    Update notes after every answered question round and every material research batch, before asking the next round.

    line 21
  5. Distinctive:

    After compaction, resume, branch switch, or suspected context loss, re-read the entire notes file before any other action.

    line 22
  6. Very specific:

    Read the recovery notes, selected spec/plan, relevant code, <work-area>/context/CONTEXT.md, any <work-area>/context/CONTEXT-MAP.md, and …

    line 26
  7. Distinctive:

    Ask one round of independent, currently answerable questions.

    line 27
  8. Distinctive:

    Do not ask a dependent question while its prerequisite is unresolved.

    line 28
  9. Distinctive:

    Stop when branches are settled, the user stops, or remaining assumptions are explicitly accepted.

    line 31
  10. Very specific:

    Challenge against the glossary: If the user's word conflicts with an existing CONTEXT.md definition, say so and ask which meaning applies.

    line 35
  11. Distinctive:

    Cross-check code: When the user states how something works and the code disagrees, surface the contradiction with file references.

    line 38

“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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Grill With Docs

Resolve user-owned design decisions through dependency-aware questions and codebase research. Persist the evolving interview state so compaction cannot strand the work. Record settled domain terms during the session and offer ADRs sparingly. Self-contained; do not require other skills to load.

Work context

Use the repository's work area, shared by all worktrees: <work-area> = ~/.agent-work/<repo-key>/, where <repo-key> is the name of the git rev-parse --path-format=absolute --git-common-dir directory, or its parent's name when that is .git/.bare, without a trailing .git. <work-area>/.repo records that directory: write it on first write; if it names another repository, stop and ask for a distinct key. Layout: work/<work-id>/notes.md for interview recovery, work/<work-id>/spec.md for requirements, work/<work-id>/plan.md for designs, shared context/ for glossaries and decisions/ for ADRs. Read selected spec/plan status, acceptance criteria, non-goals, design and testing decisions. Confirm Work-ID and reciprocal Spec/Plan links. An explicit user path, current-session confirmation, or validated reciprocal link establishes identity. Otherwise shortlist titles/statuses/summaries and ask via question tool, even for one candidate; offer Other / None—standalone discussion. Never choose by names, slugs, branch, or recency. If asking is unavailable, report the blocker. Clarify conflicts and missing links; do not pair or migrate documents automatically. A standalone discussion needs no spec/plan, but it still needs a user-confirmed Work-ID and recovery notes. Pass confirmed paths/constraints to research workers, who return missing-context questions instead of guessing.

Recovery notes: mandatory preflight

  1. Resolve or ask for the Work-ID before the first substantive interview question. This identity question is the only permitted question before recovery notes exist; do not infer the ID.
  2. Read references/notes-format.md. Inspect <work-area>/work/<work-id>/notes.md; create it if absent or merge into it without erasing unrelated history.
  3. Before the interview starts, capture all relevant information already present in the active session: objective, scope, constraints, sources, verified findings, user-reported facts, settled decisions, contradictions, open questions, and the next frontier. Summarize; never copy secrets, customer data, full transcripts, or large raw outputs.
  4. If the notes write is denied or fails, stop before the interview and report the recovery blocker. Do not rely on conversation-only state.
  5. Update notes after every answered question round and every material research batch, before asking the next round. Perform a final sync before closing or handing off to spec, plan, or implementation work.
  6. After compaction, resume, branch switch, or suspected context loss, re-read the entire notes file before any other action. If it is missing, stale, or identity-conflicted, stop and reconcile it with the user instead of continuing from memory.

Interview

  1. Read the recovery notes, selected spec/plan, relevant code, <work-area>/context/CONTEXT.md, any <work-area>/context/CONTEXT-MAP.md, and <work-area>/decisions/. Also consult existing project glossaries/ADRs as evidence; do not modify or migrate those outside <work-area> without an explicit request. Identify settled decisions, unresolved branches, factual unknowns, and terms already defined.
  2. Ask one round of independent, currently answerable questions. Use the question tool when available; give concrete options and a recommended answer with its tradeoff.
  3. Do not ask a dependent question while its prerequisite is unresolved. Wait for answers before the next round.
  4. Facts are your job: discover them from the codebase with bounded exploration. Use a worker when useful and available; otherwise inspect directly. Decisions are the user's. Blocked tools are a blocked prerequisite, not permission to guess or retry endlessly.
  5. Sync answers and evidence to recovery notes, recompute the frontier, then ask the next round. Honor a one-question-at-a-time pace or timebox.
  6. Stop when branches are settled, the user stops, or remaining assumptions are explicitly accepted.

Domain discipline (during every round)

  • Challenge against the glossary. If the user's word conflicts with an existing CONTEXT.md definition, say so and ask which meaning applies.
  • Sharpen fuzzy terms. For vague or overloaded words, propose one canonical term and name the alternatives to avoid.
  • Probe with scenarios. Invent concrete edge cases that force precise boundaries between concepts.
  • Cross-check code. When the user states how something works and the code disagrees, surface the contradiction with file references.

Paper trail

What resolvedWhere it landsWhen
Current evidence, decisions, conflicts, open questions, and next frontier<work-area>/work/<work-id>/notes.md (see notes-format.md)Before interviewing, after every round/research batch, and before handoff
A project-specific term, settled by the user's answer<work-area>/context/CONTEXT.md (see context-format.md)Immediately, inline
A decision that is hard to reverse and surprising without context and a real trade-off<work-area>/decisions/NNNN-slug.md (see adr-format.md)Only after the user accepts the offer
  • Recovery notes are a concise working ledger, not a draft spec, plan, transcript, or substitute glossary. Specs and plans remain Not created until an explicitly approved handoff creates them.
  • Except for mandatory recovery notes, create files lazily; inspect destinations before editing and preserve unrelated content. For multiple contexts, use <work-area>/context/CONTEXT-MAP.md and <work-area>/context/<context>/CONTEXT.md; ask which context when unclear. Existing maps outside this folder are evidence, not permission to write elsewhere. New artifacts stay inside <work-area> unless the user explicitly requests otherwise.
  • In each ADR, include the confirmed Work-ID and relative Spec/Plan links, or Not created for absent artifacts. Do not invent a relationship. If an accepted decision changes a selected spec/plan, flag the affected sections and request approval to reconcile them; report the mismatch until resolved. No silent implementation handoff with conflicting requirements.
  • CONTEXT.md is a glossary only: no implementation details, spec prose, or scratch notes. Terms must be specific to this domain, not general programming concepts; working detail belongs in recovery notes.
  • Offer ADRs sparingly. Most sessions produce none; that is correct.
  • Announce each write in one line (notes: updated decisions and next round, glossary: added "Invoice", adr: 0003-...). Every write goes through normal file permissions; a denied write is reported, not retried silently.

Closing

Sync recovery notes, then summarize: notes path, settled decisions, glossary entries written, ADRs written or declined, unresolved risks, and next frontier. Recommend the next step (spec, plan, or implement), pass the notes path as required handoff context, and ask for explicit approval. Do not create those artifacts or implement merely because the interview ended.

Example: tenancy model and retention policy may share a round; tenant-index design waits for tenancy. When "account" resolves to Customer, write the Customer entry with _Avoid_: account right then. "Postgres for the write model" earns an ADR offer; "name the helper parseOrder" does not.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.