The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, such as binary files, network traffic, and crash samples. This creates a potential surface for indirect prompt injection where malicious inputs analyzed by the agent (via debuggers or fuzzers) could attempt to influence the agent's logic or behavior.
Ingestion points: Target binaries, network packets captured via [[re-netcap]], and fuzzing seeds/corpora.
Boundary markers: The gateway skill does not explicitly define prompt delimiters for these inputs, relying on the orchestration of sub-skills.
Capability inventory: The skill utilizes powerful capabilities including fuzzing, symbolic execution, and constraint solving, involving tools like gdb, rr, and AFL++.
Sanitization: The skill contains explicit instructions to perform dynamic execution and reproduction within a dedicated sandbox environment ([[re-sandbox]]), which serves as a primary security control.
[COMMAND_EXECUTION]: The workflow describes the use of various command-line tools for binary analysis and vulnerability hunting, including afl-fuzz, checksec, strings, and afl-cmin. These are standard professional tools for the stated purpose of the skill.

re-vuln