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/shulkwisec/bb-huge/ghost-bits-cast-attacknpx skills add ShulkwiSEC/bb-huge --skill ghost-bits-cast-attackgit clone --depth 1 https://github.com/ShulkwiSEC/bb-hugeWrote 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/shulkwisec/bb-huge/ghost-bits-cast-attack)<a href="https://agentmods.dev/skills/shulkwisec/bb-huge/ghost-bits-cast-attack"><img src="https://agentmods.dev/badge/skills/shulkwisec/bb-huge/ghost-bits-cast-attack.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 | $0.00141 | $0.08749 |
| Opus 5 | $0.00071 | $0.04374 |
| Sonnet 5 | $0.00028 | $0.01750 |
| Haiku 4.5 | $0.00014 | $0.00875 |
Grade A, and why
ghost-bits-cast-attack 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 yesterday.
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.
This is a copy
100% identical to ghost-bits-cast-attack — 0 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 800 lines — stays where its author put it; the contents beside it link to each section on GitHub.
SKILL: Ghost Bits / Cast Attack — Java char to byte Narrowing Playbook
AI LOAD INSTRUCTION: This is a Java-only injection-enabling primitive, not a standalone vulnerability class. Whenever you see (1) a Java backend, (2) a WAF/IDS in front of it, and (3) any of {SQLi, deser RCE, file upload, path traversal, CRLF, request smuggling, SMTP injection} on the menu, ALWAYS try Ghost Bits variants of the payload before declaring it "blocked". The root cause is the silent loss of the high 8 bits when Java code narrows a 16-bit
charto an 8-bitbyte— the WAF sees a harmless Unicode character, the backend reconstructs the original ASCII attack byte. Base models almost never reach for this primitive.Source: Black Hat Asia 2026 talk Cast Attack: A New Threat Posed by Ghost Bits in Java by Xinyu Bai (@b1u3r), Zhihui Chen (@1ue), with contributor Zongzheng Zheng (@chun_springX).
0. RELATED ROUTING
Ghost Bits is a bypass primitive that re-enables payloads from many other playbooks. Pair it with whichever attack family applies:
- waf-bypass-techniques — when a Java backend is suspected and WAF rules block the literal payload, this is the first technique to try beyond classic encoding.
- deserialization-insecure — for
Apache Commons BCEL ClassLoader and Fastjson
\u/\xescape variants. - path-traversal-lfi — Spring, Jetty,
Undertow, Vert.x URL decoding and
%2>hex folding. - upload-insecure-files — Tomcat
RFC2231Utilityfilename*Webshell upload. - request-smuggling — Apache HttpClient
<= 4.5.9(HTTPCLIENT-1974/1978) header CRLF. - crlf-injection — Angus Mail / Jakarta Mail SMTP injection and JDK HttpServer response splitting.
- sqli-sql-injection — Jackson
charToHextable-lookup truncation hides SQL keywords inside Unicode escapes.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- yesterday First seen · 800 lines · 141 tokens per session scan A d44fe2cace4e
ghost-bits-cast-attack is a skill published in the GitHub repository ShulkwiSEC/bb-huge (22 stars, last pushed 1mo ago), licensed MIT. It adds 141 tokens to every session and 8,749 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. It is 100% identical to ghost-bits-cast-attack, differing in 0 lines, and is treated as a copy.
Other skills, from other repositories
bugcrowd-reporting
Bugcrowd-specific reporting tactics complementing report-writing: VRT category search-and-fallback strategy when no exact match exists, manual severity override when VRT defaults underrate impact, severity-request paragraph as first body section, OOS-clause rebuttal templates (rate limiting on auth-flow endpoints…
skill-context-detection
Auto-detect work context (Dev vs Knowledge) — use to tailor workflows based on current task type.
skill-copilot-provider
GitHub Copilot CLI as optional zero-cost provider via copilot -p programmatic mode.
race
Race condition / TOCTOU playbook — limit overrun (one-time codes used twice, gift cards spent twice), single-packet attack (last-byte sync) to force parallel processing, and state-confusion races (file upload + read, order before payment). Use when timing-sensitive logic could be abused — one-time codes, coupons/gift…
skill-security-framing
URL validation and content sanitization for untrusted sources — use when handling external input safely.
my-skill
One line on what this playbook does, then a "Use when ..." clause so the agent knows when to load it (e.g. "Use when the target exposes X / you see Y in requests"). Max 1024 chars. This description is the ONLY thing the model sees until it loads the skill — make the trigger conditions explicit, since there is no…