Grill with docs
Run a grilling session after grounding the plan in project context.
grilling is the interview workflow; this skill adds docs, code, and Obsidian
context before the questioning starts.
Workflow
- Locate project context before judging or questioning the plan. Use
grounding.md for the search targets.
- Extract the plan's nouns and verbs: product concepts, system names, state
transitions, user-visible outcomes, and implied invariants.
- Compare those terms against existing docs and code. Prefer existing
vocabulary. Flag invented terms, overloaded names, and domain gaps.
- Run
grilling: map the decision tree, ask the whole settled frontier in a
numbered round, provide a recommended answer for each question, dispatch
background fact-finding, and put decisions to the user.
- Use these prompts as useful pressure points during grilling:
- Which existing decision or ADR does this rely on?
- Which code path proves the plan is implementable?
- What behavior is the first tracer-bullet ticket?
- What assumption would make the plan fail?
- Which acceptance criterion is still vague or untestable?
- If the plan changes rendered UI, what design direction, viewport states,
and
frontend-ui-validation proof will review require?
- Apply the session behaviors in
session-behavior.md: glossary challenges,
fuzzy-language sharpening, concrete scenarios, code cross-checks, UI
readiness, context updates, and ADR offers.
- If the work is too large or foggy for one session, point the user to
break it into explicit decisions or tickets instead of forcing the whole
journey through one grill.
- Continue until the plan is clear enough to execute and the user confirms the
shared understanding.
- Return either a tightened plan with assumptions and first ticket made clear,
or a short set of blockers/questions if the plan is not ready.
Output shape
Prefer short sections:
Grounding: docs/code/notes inspected
What Holds Up: parts supported by evidence
Gaps: unclear terms, missing decisions, weak assumptions
Tightened Plan: revised plan or first ticket
Next Round: the settled frontier the user should answer next
When the plan is ready, stop asking and give the tightened plan. If the user has
not confirmed the shared understanding, make the confirmation request the next
round instead of proceeding.
Context pointers
- Use grounding.md for project context search
targets.
- Use session-behavior.md for how to
challenge terminology, code claims, scenarios, and UI readiness.
- Use context-format.md when writing or
proposing a project glossary note.
- Use adr-format.md when an ADR is genuinely
warranted.
1---2name: grill-with-docs3description: Ground a plan in repo docs, code, Obsidian notes, ADRs, specs, and tickets, then grill its decisions.4---56# Grill with docs78Run a `grilling` session after grounding the plan in project context.9`grilling` is the interview workflow; this skill adds docs, code, and Obsidian10context before the questioning starts.1112## Workflow13141. Locate project context before judging or questioning the plan. Use15 [grounding.md](references/grounding.md) for the search targets.162. Extract the plan's nouns and verbs: product concepts, system names, state17 transitions, user-visible outcomes, and implied invariants.183. Compare those terms against existing docs and code. Prefer existing19 vocabulary. Flag invented terms, overloaded names, and domain gaps.204. Run `grilling`: map the decision tree, ask the whole settled frontier in a21 numbered round, provide a recommended answer for each question, dispatch22 background fact-finding, and put decisions to the user.235. Use these prompts as useful pressure points during grilling:24 - Which existing decision or ADR does this rely on?25 - Which code path proves the plan is implementable?26 - What behavior is the first tracer-bullet ticket?27 - What assumption would make the plan fail?28 - Which acceptance criterion is still vague or untestable?29 - If the plan changes rendered UI, what design direction, viewport states,30 and `frontend-ui-validation` proof will review require?316. Apply the session behaviors in32 [session-behavior.md](references/session-behavior.md): glossary challenges,33 fuzzy-language sharpening, concrete scenarios, code cross-checks, UI34 readiness, context updates, and ADR offers.357. If the work is too large or foggy for one session, point the user to36 break it into explicit decisions or tickets instead of forcing the whole37 journey through one grill.388. Continue until the plan is clear enough to execute and the user confirms the39 shared understanding.409. Return either a tightened plan with assumptions and first ticket made clear,41 or a short set of blockers/questions if the plan is not ready.4243## Output shape4445Prefer short sections:4647- `Grounding`: docs/code/notes inspected48- `What Holds Up`: parts supported by evidence49- `Gaps`: unclear terms, missing decisions, weak assumptions50- `Tightened Plan`: revised plan or first ticket51- `Next Round`: the settled frontier the user should answer next5253When the plan is ready, stop asking and give the tightened plan. If the user has54not confirmed the shared understanding, make the confirmation request the next55round instead of proceeding.5657## Context pointers5859- Use [grounding.md](references/grounding.md) for project context search60 targets.61- Use [session-behavior.md](references/session-behavior.md) for how to62 challenge terminology, code claims, scenarios, and UI readiness.63- Use [context-format.md](references/context-format.md) when writing or64 proposing a project glossary note.65- Use [adr-format.md](references/adr-format.md) when an ADR is genuinely66 warranted.