challenge

/challenge

Two-phase challenge. Phase 1 grills, Phase 2 spawns critic-design.

Input

• $ARGUMENTS: what to challenge (proposal file path or description)
• If $ARGUMENTS is empty: stop and ask user to specify target. Silent target inference from conversation has high misfire risk.

Phase 1 Grill

Ask one question at a time. Each question must include a recommended answer presented as the top option in AskUserQuestion. Walk down each branch of the design tree, resolving dependencies one-by-one.

Rules

Rule	Detail
Code-resolvable questions	If the answer lives in the code, run ugrep / bfs instead of asking
Question budget	Cap at 7 across the phase. If hitting cap with branches still open, summarise the unresolved set and let user decide whether to continue
Stop conditions	All decision branches resolved, user signals "enough", or budget hit

Output to Phase 2

Aggregate grill findings into critic-design input schema before spawning.

Field	Source
source	"user-grill"
artifact_type	inferred from $ARGUMENTS (spec / plan / design / ADR / doc)
approach	one-line summary of the proposal core
decisions	architectural-level decisions crystallised during grill (terminology checks and scope minutiae excluded)
trade-offs	trade-offs surfaced during grill
referenced_files	files cited or read during grill

Phase 2 Devil

Step	Action
1	Compose Phase 2 input from Phase 1 aggregation + original $ARGUMENTS context
2	Spawn critic-design via Task (subagent_type: critic-design, background: false). Mention ARCHITECTURE.md or equivalent if present
3	Wait for completion, capture verdict + weaknesses

Output

Section	Content
Grill summary	Surfaced assumptions, decisions, trade-offs (one-line each)
Devil verdict	critic-design output verbatim
Actionable items	Top 3 concrete actions (keep / remove / revise)