The Agent Skills Directory

[OBFUSCATION]: The skill documents and provides a mapping for 'Ghost Bits' characters, which are Unicode characters that map to specific ASCII bytes (like '/', '.', and control characters) when processed by vulnerable Java functions. This technique is explicitly designed to bypass signature-based WAFs and IDS.- [COMMAND_EXECUTION]: Provides functional payload generators in Python and Yaklang within SKILL.md and PAYLOAD_COOKBOOK.md. These scripts are designed to create malicious payloads that exploit various vulnerabilities in Java frameworks.- [PROMPT_INJECTION]: The AI LOAD INSTRUCTION in SKILL.md contains a directive for the agent to 'ALWAYS try Ghost Bits variants' before declaring a payload blocked, which is an attempt to override the agent's default safety or logic filters when evaluating external interactions.- [INDIRECT_PROMPT_INJECTION]: The skill provides a vulnerability surface where the agent is instructed to ingest and act upon malicious patterns. Ingestion points: SKILL.md (AI LOAD INSTRUCTION). Boundary markers: None provided to separate instructions from generated payloads. Capability inventory: Functional scripts for payload generation and detailed technical mappings for multi-component exploitation. Sanitization: No validation or sanitization is performed on the generated payloads.- [REMOTE_CODE_EXECUTION]: The skill provides detailed instructions, specific CVE references (e.g., CVE-2024-36401, CVE-2022-22965), and payload templates to facilitate remote code execution on Java backends using the described narrowing primitive.

ghost-bits-cast-attack