goncho-grill-me

name	goncho-grill-me
description	Stress-test Goncho-specific implementation plans before coding. Use when asked to grill a Goncho roadmap slice, validate readiness for TDD, challenge a memory-system design, or find gaps against Goncho docs/metaanalysis.

Goncho Grill Me

Quick start

Inspect relevant Goncho docs/code first, then ask one hard question at a time with a recommended answer.

Use generic grill-me for non-Goncho plans. Use this skill only when the plan must fit Goncho's memory architecture, roadmap, tests, or metaanalysis.

Workflow

1. Ground the plan
   • Identify the TODO/README/docs/metaanalysis capability.
   • Inspect current files/tests instead of asking questions the repo can answer.
   • Name the current implementation status and public contract/API surface.
2. Challenge the design
   • Capability fit: what gap closes?
   • Trust model: evidence, scope, time, confidence, authority, lifecycle.
   • User-visible behavior: search, context, review, audit, tool, HTTP, or CLI output.
   • Failure mode: stale, conflicting, leaking, noisy, or repeated-bad-memory case.
   • TDD proof: first failing test and why it fails now.
   • Minimal slice: smallest vertical behavior.
   • Non-goals: tempting phase work excluded.
3. Stop before implementation
   • If implementation is approved, switch to goncho-tdd-implementation.

Question format

Question N: <specific challenge>
Why it matters: <risk or design dependency>
Recommended answer: <concrete default>

Ask only one decision-driving question at a time.

Skill contract

Entry protocol

• Trivial: if the plan is already concrete, summarize readiness and the first failing test.
• Medium ambiguity: propose the likely slice and ask the single hard owner decision.
• High ambiguity/risk: stop if ownership, trust boundary, or destructive/public behavior is unclear.

Topology check

• State/ownership: memory data, evidence, lifecycle, host authority, and adapter boundaries.
• Feedback/validation: exact RED test, user-visible output, and full verification command.
• Blast radius: public API/tool behavior, migrations, benchmarks, docs, server/team mode.
• Timing/ordering: imports, recalls, review, supersession, concurrency, release sequencing.

Verification gate

A grill session is done when it names the chosen slice, first failing test, likely files, risks/non-goals, and asks for approval before TDD.

Red lines

• Do not implement during grilling.
• Do not ask broad questionnaires; inspect first and ask one hard question.
• Do not accept schema-only completion for user-visible features.
• Do not expand one slice into a whole roadmap phase.

Output contract

End with: chosen slice, first failing test, likely files, risks/non-goals, and explicit ask to start TDD.