Bug Magnet Data

Curated edge case test data for boundary testing, verification scripts, and test generation. 50+ years of testing wisdom distilled into small, high-signal collections organized by data type.

Core Principle: Curation beats generation. 50 well-chosen edge cases find more bugs than 10,000 random inputs.

---

When to Use This Skill

Load this skill when the consumer request matches ANY of these patterns:

Consumer	Trigger	Usage
test-audit	Step 7 (edge case gap detection)	Identify missing boundary test coverage
bulwark-verify	Generating verification scripts	Inject edge cases into test scenarios
bulwark-fix-validator	Validating a fix	Test fix against boundary conditions

DO NOT use for:

• Encrypted/compressed data (edge cases won't penetrate wrapping)
• Pure unit tests with fully mocked dependencies (edge cases need real execution)
• Performance/load testing (use dedicated load testing tools)

---

Pre-Flight Gate (BLOCKING)

STOP. Before providing ANY edge case data, you MUST follow the three-phase workflow.

This skill provides curated data through a deterministic workflow. You must execute all phases.

What You MUST Do

1. Phase 1: Component Detection - Identify component type and load context file
2. Phase 2: Data Loading - Load T0 + T1 data files (REQUIRED), T2 if specified by context
3. Phase 3: Edge Case Application - Apply edge cases and report what was loaded

What You MUST NOT Do

• Do NOT generate edge cases from your own knowledge - use the curated data files
• Do NOT skip loading context files - they determine which categories apply
• Do NOT skip the safety filter - patterns marked safe_for_automation: false must be excluded
• Do NOT return partial data - all applicable tiers must be loaded

Why This Matters

The curated data exists because:

• Curation beats generation - 50 well-chosen edge cases find more bugs than 10,000 random inputs
• Reproducibility - Same component type = same edge cases every time
• Safety - Destructive patterns are explicitly marked and filtered

If you find yourself thinking "I know some good edge cases" - STOP. Use the data files.

Completion Checklist

Before returning to consumer, verify ALL items:

• Phase 1: Component type detected
• Phase 1: Context file loaded for component type
• Phase 2: T0 data files loaded (boundaries, booleans, collections)
• Phase 2: T1 data files loaded (unicode, special-chars, injection, special numbers)
• Phase 2: T2 data files loaded (if specified by context file)
• Phase 2: Safety filter applied (excluded manual_only and safe_for_automation: false)
• Phase 3: Edge cases applied to test/verification scenario
• Phase 3: Report includes categories loaded and patterns excluded

Do NOT return to consumer until all checkboxes can be marked complete.

---

Dependencies

This skill provides data files and context guidance. Understanding what to load ensures deterministic execution.

Category	Files	Requirement	When to Load
Context files	context/{component-type}.md	REQUIRED	Always load for detected component type
T0 data (boundaries)	data/strings/boundaries.yaml, data/numbers/boundaries.yaml, data/booleans/boundaries.yaml, data/collections/arrays.yaml	REQUIRED	Every edge case injection
T1 data (common)	data/strings/unicode.yaml, data/strings/special-chars.yaml, data/strings/injection.yaml, data/numbers/special.yaml	REQUIRED	Most edge case injections
T2 data (context-specific)	data/dates/*.yaml, data/encoding/*.yaml, data/formats/*.yaml, data/concurrency/*.yaml	CONDITIONALLY REQUIRED	If context file specifies → MUST load
Language-specific	data/language-specific/{language}.yaml	CONDITIONALLY REQUIRED	If testing language-specific behavior → MUST load
External references	references/external-lists.md	REQUIRED	For source attribution and update checking

Fallback behavior:

• If component type detected → Loading context/{type}.md is REQUIRED
• If context file specifies a category → Loading that category is REQUIRED
• If a referenced file is missing → Note in output, continue with available data

---

Data Tiers

Tier	Categories	When to Load
T0 (Always)	Boundaries (empty/single/max), Null handling	Every edge case injection
T1 (Common)	Basic injection, Unicode basics, Numeric edges	Every edge case injection
T2 (Context)	Date/time, Encoding, Formats, Concurrency	When context file specifies
T3 (Manual)	Patterns marked manual_only: true	NEVER for automated runs

Safety Filtering: Patterns with safe_for_automation: false or manual_only: true MUST be excluded from automated test runs.

---

Three-Phase Workflow

CRITICAL: All three phases are REQUIRED. Do not skip any phase.

Phase 1: Component Detection (Deterministic)
├── Identify component type from code under test
├── Map to context file: context/{cli-args|http-body|file-contents|db-query|process-spawn}.md
└── Load context file → get applicable categories list

Phase 2: Data Loading (Deterministic)
├── Load T0 data files (REQUIRED - always)
├── Load T1 data files (REQUIRED - always)
├── Load T2 data files specified by context file (CONDITIONALLY REQUIRED)
├── Load language-specific file if applicable (CONDITIONALLY REQUIRED)
└── Apply safety filter: exclude patterns with safe_for_automation: false

Phase 3: Edge Case Application
├── Inject loaded edge cases into test scenarios
├── Report which categories were loaded
└── Report any patterns excluded due to safety filtering

---

Component Type Detection

Map code under test to component type. Detection determines which context file to load.

Code Pattern	Component Type	Context File
CLI argument parsing, process.argv, argparse	CLI	context/cli-args.md
HTTP request/response, req.body, fetch, axios	HTTP	context/http-body.md
File I/O, fs.read, open(), file parsing	File	context/file-contents.md
Database queries, SQL, ORM operations	Database	context/db-query.md
Child process, spawn, exec, subprocess	Process	context/process-spawn.md

If multiple types apply: Load context files for each applicable type.

---

Category Reference

Strings (T0/T1)

File	Contents	Bugs Caught
strings/boundaries.yaml	Empty, single char, long strings, whitespace	NullPointerException, buffer overflow, off-by-one
strings/unicode.yaml	Multi-byte, normalization, emoji, RTL	Encoding errors, length calculation bugs
strings/special-chars.yaml	Quotes, escapes, control characters	Escape sequence handling, delimiter confusion
strings/injection.yaml	SQL, XSS, command injection, path traversal	Security vulnerabilities

Numbers (T0/T1)

File	Contents	Bugs Caught
numbers/boundaries.yaml	0, -1, 1, MAX_INT, MIN_INT	Integer overflow/underflow, off-by-one
numbers/special.yaml	NaN, Infinity, -0	Special value handling, NaN propagation
numbers/precision.yaml	0.1+0.2, large/small floats	Floating point comparison failures

Booleans (T0)

File	Contents	Bugs Caught
booleans/boundaries.yaml	true, false, null, truthy/falsy	Null reference, truthy/falsy confusion

Collections (T0)

File	Contents	Bugs Caught
collections/arrays.yaml	Empty, single, large, nested, sparse	Index out of bounds, empty collection crashes
collections/objects.yaml	Empty, nested, circular, prototype pollution	Null reference, prototype pollution

Dates (T2)

File	Contents	Bugs Caught
dates/boundaries.yaml	Epoch, Y2K38, leap year	Y2K38 overflow, leap year bugs
dates/timezone.yaml	DST transitions, UTC offsets	DST errors, timezone conversion
dates/invalid.yaml	Feb 30, invalid formats	Date parsing failures

Encoding (T2)

File	Contents	Bugs Caught
encoding/charset.yaml	ASCII, UTF-8, BOM	Encoding detection, mojibake
encoding/normalization.yaml	NFC, NFD, overlong	Normalization mismatches

Formats (T2)

File	Contents	Bugs Caught
formats/email.yaml	Valid/invalid patterns	Overly strict/lenient validation
formats/url.yaml	Valid/invalid patterns	URL parsing errors
formats/json.yaml	Valid/invalid patterns	JSON parsing errors

Concurrency (T2)

File	Contents	Bugs Caught
concurrency/race-conditions.yaml	Double submit, concurrent edit	Race conditions, lost updates
concurrency/state-machines.yaml	Invalid transitions	State corruption

Language-Specific (Conditional)

File	Contents	Bugs Caught
language-specific/javascript.yaml	== vs ===, truthy/falsy	Type coercion bugs
language-specific/python.yaml	None vs False, mutable defaults	Python-specific gotchas
language-specific/rust.yaml	Ownership, borrowing	Memory safety issues

---

Data File Format

All data files use this YAML structure:

metadata:
  version: "1.0.0"
  last_updated: "2026-02-01"
  source_urls: []

category: strings
subcategory: boundaries
tier: T0

values:
  identifier:
    value: "actual value"
    bugs_caught: ["Bug type 1", "Bug type 2"]
    safe_for_automation: true
    manual_only: false

Safety flags to check:

• safe_for_automation: false → Exclude from automated runs
• manual_only: true → NEVER include in automated runs

---

Integration Examples

test-audit (Step 7)

1. Detect component type from test file
2. Load context file for component type
3. Load T0 + T1 data files
4. Load T2 files specified by context
5. Compare test coverage against loaded edge cases
6. Report missing edge case coverage

bulwark-verify

1. Detect component type from code under verification
2. Load context file for component type
3. Load T0 + T1 data files
4. Load T2 files specified by context
5. Filter out patterns with safe_for_automation: false
6. Inject applicable edge cases into verification script

bulwark-fix-validator

1. Detect component type from fix
2. Load context file for component type
3. Load T0 + T1 data files
4. Test fix against loaded edge cases
5. Report any edge cases that break the fix