可以将 grill-me 加入到 Trellis 的 workflow 中使用

这个方法是站内佬友的方法,将下面的 diff 给 ai 让他修改
11a12
> 6. **Grill before execution** — stress-test ambiguous or risky plans before implementation
102c103
< Phase 1: Plan → figure out what to do (brainstorm + research → prd.md)
---
> Phase 1: Plan → figure out what to do (brainstorm + grill + research → prd.md)
141a143
> | Wants to stress-test a plan/design, says "grill me", asks for ruthless review, or a draft PRD still has unresolved design decisions | `grill-me` |
155a158
> | Wants to stress-test a plan/design, says "grill me", asks for ruthless review, or a draft PRD still has unresolved design decisions | `grill-me` |
226c229,245
< Return to this step whenever requirements change and revise `prd.md`.
---
> After the first coherent draft of `prd.md` exists, decide whether a `grill-me` pass is required before leaving Phase 1.1.
>
> Load `grill-me` when any of these are true:
> - The user explicitly says "grill me", asks to be stress-tested, or wants ruthless review
> - The plan has architecture, API, data model, product-scope, migration, security, performance, or cross-platform decisions
> - The PRD contains alternatives, TODOs, assumptions, hidden dependencies, or vague acceptance criteria
> - Implementation would be expensive to unwind if the plan is wrong
>
> When `grill-me` is active:
> - Ask exactly one question at a time
> - For each question, include the recommended answer
> - Walk the decision tree branch-by-branch and resolve dependencies before moving on
> - If a question can be answered by reading the codebase, specs, docs, or research artifacts, explore those instead of asking the user
> - After each user answer, immediately update `prd.md` and, for technical design details, `info.md`
> - Stop only when the remaining plan is specific enough for Phase 1.3 context curation and Phase 2 implementation
>
> Return to this step whenever requirements change and revise `prd.md`. If the change reopens design decisions, run another focused `grill-me` pass.
313a333
> | `grill-me` pass completed when the task is complex, risky, or explicitly requested | ✅ when applicable |
519c539
< Flow: load `trellis-brainstorm` skill → it creates the task via `python3 ./.trellis/scripts/task.py create` and drives requirements Q&A. For research-heavy work (tool comparison, docs, cross-platform survey), spawn `trellis-research` sub-agents via Task tool — NEVER do 3+ inline WebFetch/WebSearch/`gh api` calls in the main conversation.
---
> Flow: load `trellis-brainstorm` skill → it creates the task via `python3 ./.trellis/scripts/task.py create` and drives requirements Q&A. Once a draft plan exists, load `grill-me` before implementation if the user asks to be grilled or the plan still has unresolved design decisions. For research-heavy work (tool comparison, docs, cross-platform survey), spawn `trellis-research` sub-agents via Task tool — NEVER do 3+ inline WebFetch/WebSearch/`gh api` calls in the main conversation.
524c544
< Complete prd.md via trellis-brainstorm skill; then run task.py start.
---
> Complete `prd.md` via `trellis-brainstorm`. If the user asked to be grilled or the plan is complex/risky, run `grill-me` one question at a time until key decisions are resolved, updating `prd.md` / `info.md` after each answer. Then run `task.py start` if the task is not already active.