Decision Grill 🔥

Grill every technical decision until it's bulletproof. No "I think this is fine."

Why This Skill Exists

AI agents love quick answers. Quick answers hide:

• Unexplored alternatives
• Hidden trade-offs
• Assumptions not validated
• Edge cases not considered

Result: "It should work" → Production incident.

The 23.5% incident spike from AI-generated code starts here.

The 4-Phase Grill Protocol

Phase	Duration	What AI Must Prove
1. Options	5 min	List ALL approaches (min 3), explain dismissed ones
2. Trade-offs	10 min	For EACH option: easy, hard, future problems, critics
3. Edge Cases	10 min	Failure cases, weird inputs, scale limits, dependencies
4. Decision Tree	5 min	Complete tree with reasons, risks, monitoring

Red flags trigger GRILL HARDER:

• Only 1-2 options named
• Trade-offs seem too mild
• No failure cases found
• Decision tree has "?" branches

Example Grill Session

Without grill: "PostgreSQL is great. Use that."

With grill:

🔥 DECISION GRILL ACTIVATED 🔥

Phase 1: Options
├─ PostgreSQL ✓ (considering)
├─ MongoDB (dismissed: no ACID needed here)
├─ SQLite (dismissed: not for multi-server)
├─ DynamoDB (dismissed: don't want AWS lock-in)
└─ Redis (dismissed: not persistent enough)

Phase 2: Trade-offs for PostgreSQL
├─ Makes easy: SQL, joins, relations, mature ecosystem
├─ Makes hard: Horizontal scaling (need Citus or sharding)
├─ Future problems: Connection pool limits at scale
└─ Who hates it: ORM zealots, NoSQL fans

Phase 3: Edge cases
├─ Fails when: >10K concurrent writes without partitioning
├─ Breaks on: Extremely nested JSON queries
├─ Scale limit: ~1TB per table before performance dips
└─ Dependency risk: PostGIS extension for geo queries

Phase 4: Decision Tree
Decision: PostgreSQL
├─ Why not MongoDB? Need transactions for orders
├─ Why not SQLite? Planning multi-server deployment
├─ Accepted: Manual sharding later if needed
├─ Risks: Connection pool exhaustion at 10K+ users
└─ Monitor: Query latency, connection count

✓ GRILL COMPLETE

Grill Intensity Levels

Level	Use Case	Proof Required
Light	Minor refactor, format choice	2 options, basic trade-offs
Medium	New feature, library choice	3 options, edge cases, trade-offs
Heavy	Architecture, security, data model	5 options, failure analysis, full tree
Extreme	Production-critical, no rollback	Everything + team review + rollback plan

Quick Start

# For OpenClaw
clawhub install decision-grill

# For Claude Code
mkdir -p ~/.claude/skills/decision-grill
curl -o ~/.claude/skills/decision-grill/SKILL.md \
  https://raw.githubusercontent.com/aptratcn/skill-decision-grill/main/SKILL.md

Market Validation

Inspired by mattpocock/skills/grill-me trending on GitHub — users want AI to stop giving quick answers and start thinking through decisions.

mattpocock's grill-me: "Get relentlessly interviewed about a plan or design until every branch of the decision tree is resolved."

decision-grill: Same philosophy, but structured for technical decisions with:

• 4-phase protocol
• Intensity levels
• Decision tree output format

Related Skills

• evr-framework — Verify before claiming complete
• cognitive-debt-guard — Prevent AI code incidents
• systematic-debugging — Root cause analysis

---

Stop "I think this is fine."
Start "I've grilled this from every angle."