scenario

Activate When

• /godmode:scenario, "edge cases", "failure modes"
• "what could go wrong?", "break this", "stress test"
• Before shipping features with user input or APIs
• After build when no edge case coverage exists

Workflow

1. Read the Design

Read spec, plan, and code. Trace the data flow:

Input -> Validation -> Transform -> Logic -> Persist
  -> Side Effects -> Response

For each stage list:

• External calls: API endpoints, DB queries, file I/O
• State mutations: created, updated, deleted
• Side effects: emails, webhooks, cache invalidation
• Trust boundaries: where untrusted data enters

Print: [scenario:read] Feature: {name} | {N} calls | {N} mutations | {N} side effects | {N} boundaries

2. Identify Feature Scope

Entry points: API routes, CLI, UI events, cron, queues
Dependencies: services, DBs, caches, third-party APIs
Data shapes: input schemas, output schemas, models

3. Explore All 12 Dimensions

For EACH dimension, generate 2-5 specific scenarios. Every scenario MUST reference a code path (file:line).

#	Dimension	What to explore
1	Invalid Input	SQL injection, XSS, >64KB strings
2	Boundary	0, -1, MAX_INT+1, empty, null
3	Concurrency	Race conditions, double-submit
4	Network	Timeout, partial response, DNS
5	Data Integrity	Orphans, constraint violations
6	Auth	Expired tokens, privilege escalation
7	Time	Timezone, DST, leap seconds, TTL
8	Scale	10x load, large payloads, N+1
9	Failure	Crash mid-write, OOM, disk full
10	Migration	Schema change, data backfill
11	User Error	Double-click, back button, paste
12	Config	Missing env var, wrong region

IF dimension truly N/A: one-sentence justification. "N/A" alone is not acceptable.

4. Score Every Scenario

• Likelihood (L): 1-5 (impossible to inevitable)
• Impact (I): 1-5 (cosmetic to system down)
• Score = L x I

CRITICAL: Score >= 20 (e.g., L=5 x I=4)
HIGH:     Score 12-19
MEDIUM:   Score 6-11
LOW:      Score 1-5
Minimum valid score is 1. Zero is invalid.

IF scenario covers <80% of edge cases: add more. WHEN risk score >7: escalate to team lead.

5. Sort and Prioritize

Sort by score descending. Within same score:

1. Data loss / security breach first
2. Most users affected second
3. Easiest reproduction third

DIMENSION     | SCENARIO              | L | I | SCORE
Invalid Input | SQL injection search  | 4 | 5 | 20
Concurrency   | Race on balance       | 3 | 5 | 15

6. Generate Tests for HIGH+ Scenarios

For every scenario scored 12+:

# Detect test framework
ls package.json pytest.ini Cargo.toml 2>/dev/null
cat package.json 2>/dev/null | grep -E "vitest|jest|mocha"

# Run generated tests
npx vitest run tests/scenarios/ 2>&1

• Use project's test framework (auto-detect)
• ARRANGE / ACT / ASSERT structure
• Real code paths, no stubs, no TODOs
• Save to tests/scenarios/{feature}.scenario.test.{ext}

7. Verify Tests Run

test_cmd -- tests/scenarios/{feature}.scenario.test.{ext}

• IF test fails to compile: fix syntax immediately
• IF test fails because real bug: flag as BUG_FOUND
• IF test passes: edge case is mitigated

8. Output Summary

Scenario Analysis: {feature}
{total} scenarios across 12 dimensions
{critical} CRITICAL | {high} HIGH | {medium} MEDIUM
{tests} tests generated | {bugs} bugs found

TSV Logging

File: .godmode/scenario-log.tsv

timestamp	feature	dimension	scenario	L	I	score	severity	test_file	code_ref	status

Status: TEST_PASS, TEST_FAIL, BUG_FOUND, NO_TEST

Quality Targets

• Target: >90% of critical user paths covered
• Scenario execution: <60s per end-to-end scenario
• Target: 0 flaky scenarios (quarantine >1% failure rate)

Hard Rules

1. Explore all 12 dimensions. Skip needs justification.
2. Every scenario needs file:line or spec heading.
3. Score every scenario: L(1-5) x I(1-5). No zeros.
4. HIGH+ scenarios must have runnable tests.
5. Tests use project's actual framework, real paths.
6. Guard: test_cmd && lint_cmd.
7. On failure: git reset --hard HEAD~1.
8. Never ask to continue. Loop autonomously.

Keep/Discard Discipline

KEEP if: test is valid AND uses real code paths
  AND follows ARRANGE/ACT/ASSERT
DISCARD if: syntax errors after 2 fixes
  OR references nonexistent code
  On discard: log as NO_TEST. No stub tests.

Stop Conditions

STOP when FIRST of:
  - All 12 dimensions explored AND HIGH+ have tests
  - 3 consecutive dimensions produce 0 HIGH+ scenarios
  - >5 test generation failures with no output