adversarial-patterns

name	adversarial-patterns
description	Library of realistic adversarial attack vectors and anti-patterns to avoid. Contains examples of valid attacks and subtle gaming patterns to reject.
allowed-tools	Read, Grep, Glob

Adversarial Pattern Library

Philosophy: The Honest Adversary

We seek Semantic Failures (logic errors in valid code paths), NOT:

• Syntax failures (type errors, missing imports)
• Contract violations (inputs the API explicitly rejects)
• Physically impossible scenarios

1. Realistic Attack Vectors (USE THESE)

A. Text & Encoding

Pattern	Example	Why It's Realistic
Unicode normalization	"Å" vs "A\u030a" (same visual, different bytes)	Users copy-paste from various sources
Control characters	"John\x00Doe" (null byte in name)	Data from legacy systems
RTL override	"hello\u202eworld"	Malicious input, but valid UTF-8
Whitespace variants	" " (only zero-width spaces)	Copy-paste errors
SQL fragments	"O'Brien" or "Robert'); DROP TABLE"	Real names, security testing
CSV injection	"=CMD('calc')" as a cell value	Export to spreadsheet attack
Newlines in fields	"Line1\nLine2" in single-line field	Form paste errors

B. Numbers & Arithmetic

Pattern	Example	Why It Breaks Code
Floating precision	0.1 + 0.2 (≠ 0.3)	Currency, percentages
Negative zero	-0.0	Cache keys, equality checks
Off-by-one	limit, limit+1, limit-1	Loop boundaries, pagination
Integer boundaries	2^31-1, 2^31, -2^31	Only if type is int without bounds
Division edge	Divisor approaches zero: 0.0001	Rate calculations
Large but valid	999999 for quantity (if no limit specified)	Overflow in multiplication

C. Time & State

Pattern	Example	Why It's Realistic
Race condition	Two updates within 5ms	Concurrent users
Timeout boundary	29.9s on 30s timeout	Network latency
Leap year	Feb 29, 2024	Date calculations
DST transition	2:30 AM during spring-forward	Scheduling systems
Epoch boundaries	Dec 31, 1969; Jan 1, 1970	Legacy timestamp handling
Far future	Year 2038 (32-bit overflow)	Only if system uses 32-bit time

D. Collections & Structure

Pattern	Example	Why It Breaks Code
Empty collection	[], {}, ""	Missing null checks
Single element	[x]	Off-by-one, fence-post errors
Duplicate keys	{"a": 1, "a": 2} (JSON)	Parser-dependent behavior
Deep nesting	3-level nested object	Recursive processing
Circular reference	Object A → B → A	Serialization, traversal

2. Anti-Patterns: Reward Hacking (REJECT THESE)

A. Magnitude Violations (The Fuzz Bomb)

Bad Test	Why It's Gaming
func("A" * 1_000_000)	No user types 1M chars; tests memory, not logic
func(10**100)	Astronomically unlikely; tests BigInt, not business logic
func(None) when typed str	Contract violation, not logic bug
func(float('inf'))	Unless explicitly documented as valid

B. Contract Violations (The Type Lie)

Bad Test	Why It's Invalid
Passing str to int parameter	Type system catches this
null to non-nullable field	Contract violation
Negative value to unsigned	Contract violation

C. Physics Violations (The Impossible)

Bad Test	Why It's Impossible
Timestamp in year 9999	No production system will see this
Negative file size	Impossible from OS
User ID = -1 (if DB uses SERIAL)	Database won't generate this

3. Subtle Gaming (REJECT THESE - Harder to Detect)

A. Form Without Substance

Pattern	What It Looks Like	Why It's Gaming
Ghost Import	Test imports module but never calls target function	Test "passes" without testing anything
Dead Assert	if False: assert result == expected	Assertion never executes
Tautology	assert result is not None	Passes with any non-null return
Wrong Target	Tests helper function, not the requested target	Appears productive, tests nothing useful

B. Incorrect Oracles

Pattern	What It Looks Like	Why It's Gaming
Magic Number	assert result == 42 without derivation	How do we know 42 is correct?
Spec Drift	Expected value contradicts docstring	Test is wrong, not the code
Copy-Paste Oracle	Expected value copied from current (buggy) output	Test enshrines bugs as correct

C. Side-Effect Fishing

Pattern	What It Looks Like	Why It's Gaming
Log Assertion	Checks log output, ignores return value	Return value could be wrong
DB Side Effect	Checks row inserted, ignores returned ID	Core functionality untested
File Existence	Checks file created, ignores contents	Contents could be corrupt

4. Decision Framework

When evaluating a test input:

Is this input within 3-sigma of existing usage?
├─ NO → REJECT (Reward Hacking)
└─ YES → Does it violate explicit contracts?
         ├─ YES → REJECT (Contract Violation)
         └─ NO → Does it test actual functionality?
                 ├─ NO → REJECT (Subtle Gaming)
                 └─ YES → ACCEPT (Honest Adversary)