The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains instructions to execute shell commands for environment setup, package installation, and running the fuzzer. These include commands like export ASAN_OPTIONS, gem install, and ruby execution. These are standard operations required for the skill's primary purpose of security testing.
[EXTERNAL_DOWNLOADS]: The instructions reference external resources including the official GitHub repository for the project and the RubyGems registry for installing the ruzzy package. These references target the author's official distribution channels.
[DYNAMIC_EXECUTION]: The skill utilizes LD_PRELOAD to inject AddressSanitizer (ASan) and UndefinedBehaviorSanitizer (UBSan) libraries into the Ruby process at runtime. This is the standard and necessary mechanism for coverage-guided fuzzing and memory safety detection. Additionally, the skill involves native code compilation using clang during gem installation.
[INDIRECT_PROMPT_INJECTION]: The skill describes a fuzzing process which, by definition, involves ingesting untrusted data into a target function.
Ingestion points: The test_one_input lambda in the harness scripts (e.g., test_harness.rb, fuzz_msgpack.rb) receives raw byte data from the fuzzer engine.
Boundary markers: The skill documentation explicitly instructs users to catch Ruby exceptions (rescue Exception) to prevent the fuzzer from exiting prematurely on expected malformed inputs, effectively isolating interpreter-level errors from memory safety issues.
Capability inventory: The skill has the capability to execute code via the Ruby interpreter and native C extensions using the provided harnesses.
Sanitization: The provided harness templates include error handling blocks to safely process malformed input data during the fuzzing loop.

ruzzy