Adversarial

Adversarial Thinking

Integration: Load via @Skill:adversarial.md in DECISION LOOP step 5
RTK Integration: Supports @hooks-session-start.md, @PreCompact.md

Overview

Adversarial thinking for security research, vulnerability assessment, and bypass testing. Requires attacker mindset while maintaining defensive best practices.

When to Activate

Use this skill when tasks involve:

• Security testing or penetration testing
• Bypassing filters, validators, or rate limiters
• Vulnerability research (ethical/security contexts only)
• Understanding attack vectors for defense planning
• Testing input sanitization and validation

Detection Keywords

"bypass", "break", "exploit", "circumvent", "evade", "filter", "XSS", "injection", "sanitize", "vulnerability", "penetration", "audit"

Protocol Integration

DECISION LOOP Position

This skill applies at step 5 of the DECISION LOOP:

1. CLASSIFY  -> complexity? backup needed? tools?
2. PROTECT   -> cp file file.bak (for configs, DBs)
3. MEMORY    -> query relevant context + past failures
4. AGENTS    -> check overlaps (if multi-agent)
5. SKILLS    -> @Skill:adversarial.md for domain-specific guidance
6. WORK      -> implement (ALWAYS use worktree for ANY file changes)
7. REVIEW    -> self-review diff before testing
8. TEST      -> completion gates pass
9. LEARN     -> store outcome in memory

Required Pre-Checks

Before applying adversarial patterns:

• MANDATORY: Worktree created (uap worktree create <slug>)
• Schema diff gate completed (if tests involved)
• Environment check performed
• Memory queried for relevant past failures
• Ethical guidelines confirmed (security research only)

Attack Vector Enumeration

Systematic Checklist

1. Case variation: <ScRiPt>, SELECT vs select
2. Encoding: URL encode (%3C), HTML entities (<), Unicode (\u003c)
3. Null bytes: %00 to truncate strings
4. Double encoding: %253C -> %3C -> <
5. Context breaking: Close existing tag/string before injecting
6. Event handlers: onerror=, onload=, onfocus=
7. Alternative syntax: Template literals, eval(), Function()
8. Whitespace tricks: Tabs, newlines, zero-width characters
9. Protocol handlers: javascript:, data:, vbscript:
10. Mutation XSS: HTML parser normalization differences

Testing Protocol

• Test EACH vector until one works -- do not stop at first failure
• Document all vectors tried and results
• Use established libraries when building defenses (DOMPurify, bleach)

Defensive Patterns

CRITICAL: When building defense, ALWAYS use allow-list approach

# ❌ NEVER use blacklist/regex to filter malicious patterns
if re.search(r'<script|javascript:', user_input):  # WRONG

# ✅ Use established security libraries
import bleach
clean = bleach.clean(user_input, tags=[], attributes={})

# ✅ Or parameterized queries for SQL
cursor.execute("SELECT * FROM users WHERE id = ?", (user_id,))

UAP Protocol Compliance

Completion Gates Checklist

Before reporting task complete:

[x] Schema diffed against test expectations
[x] Tests: X/Y (must be 100%, run 3+ times)
[x] Outputs verified: ls -la
[x] Worktree created and PR prepared
[x] MANDATORY cleanup after PR merge

Verifier-First Integration

If tests are involved:

1. Schema Diff Gate: Compare output schema against test expectations BEFORE running verifier
2. Environment Check: python3 --version; which pip3 gcc make node npm
3. Run Verifier: After each change, minimum 3 runs with TESTS output format

Memory Integration

Store Lessons Learned

uap memory store "Adversarial pattern: <pattern_name> bypassed <defense_type>" \
  --tags adversarial,<category> --importance 8

Query Relevant Context

uap memory query "adversarial bypass <target_type>"

Common Pitfalls

1. Never skip worktree: ALWAYS use uap worktree create <slug> before any file changes
2. Never claim security via regex: Use established libraries (bleach, DOMPurify)
3. Never test on production: Always use isolated environments
4. Never forget cleanup: Run uap worktree cleanup <id> after PR merge

Success Criteria

• All tests pass (100%, 3+ runs)
• Schema verified against test expectations
• Worktree cleanup completed
• Lessons stored in memory
• No debug artifacts left behind
• Ethical guidelines followed

References

• CLAUDE.md v2.3.0: Universal Agent Patterns (P1-P39)
• OWASP Top 10: https://owasp.org/www-project-top-ten/
• Pattern P10: Whitelist-First Sanitization
• Pattern P20: "bypass/break/exploit" Attacker Mindset

Last Updated: 2026-03-09
Compatibility: CLAUDE.md v2.3.0+
RTK Includes: @hooks-session-start.md, @PreCompact.md