Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/jimmc414/claude-code-plugin-marketplace/adversarial-patternsnpx skills add jimmc414/claude-code-plugin-marketplace --skill adversarial-patternsgit clone --depth 1 https://github.com/jimmc414/claude-code-plugin-marketplaceWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/jimmc414/claude-code-plugin-marketplace/adversarial-patterns)<a href="https://agentmods.dev/skills/jimmc414/claude-code-plugin-marketplace/adversarial-patterns"><img src="https://agentmods.dev/badge/skills/jimmc414/claude-code-plugin-marketplace/adversarial-patterns.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00032 | $0.01407 |
| Opus 5 | $0.00016 | $0.00704 |
| Sonnet 5 | $0.00006 | $0.00281 |
| Haiku 4.5 | $0.00003 | $0.00141 |
Grade A, and why
adversarial-patterns scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 119 lines — stays where its author put it; the contents beside it link to each section on GitHub.
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 |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 6d ago First seen · 119 lines · 32 tokens per session scan A a5acb91df19c
adversarial-patterns is a skill published in the GitHub repository jimmc414/claude-code-plugin-marketplace (4 stars, last pushed yesterday), licensed MIT. It adds 32 tokens to every session and 1,407 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
audit-progressive-disclosure
Read-only progressive-disclosure audit for agent-facing instruction markdown. Grades every target against a three-tier load-cost model (always-loaded / invocation-loaded / on-demand) and classifies seven finding shapes in two lanes: split opportunities (oversize vs tier-calibrated Anthropic-prescribed caps…
quiz-me
Post-work comprehension check: after a change is complete, generate a self-contained HTML report of what was done (context, intuition, decisions) with a quiz at the bottom that you answer. Verifying the HUMAN absorbed the work, not the artifact. Non-gating by default; the quizpolicy userConfig tunes offer cadence.…
generate
Build a source-backed AI industry briefing from official vendor publications, configured RSS feeds, GitHub releases, reputable secondary reporting, and user-supplied URLs. Use when: 'ai briefing', 'ai news', 'what's new in AI', 'catch me up on AI', 'prep for AI meeting', 'AI roundup', or 'generate AI slides'.
changelog
Ingest Claude Code changelog entries and integrate them into the current repo. Fetch (read-only display), diff (impact analysis, no edits), status (applied versions), and apply (full integrate pipeline, explicit user intent only). Use when: 'new cc version', 'what changed in claude code', 'apply changelog', a new CC…
eli5
Dead-simple VISUAL explainer. Produces a visual HTML explainer that assumes zero prior knowledge: one idea per diagram, minimal text. Works on a codebase object (a module, a tradeoff, an incident) or a general concept, and grounds in the real artifact before drawing anything. Use when: 'ELI5', 'explain like I'm five'…
use
Reach Anthropic's first-party playground capability in one step: verify the upstream playground plugin is installed, invoke its skill for interactive single-file HTML explorers (controls beside a live preview, output is a prompt you paste back), or emit the exact install commands when it is absent. Use when: 'make me…