The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading specialized security research tools from GitHub, specifically ctgrind and dudect. These are established tools in the security community for identifying timing leaks in cryptographic code.
[COMMAND_EXECUTION]: The workflow involves executing local commands for code analysis, including ripgrep for source searching, objdump for disassembling compiled binaries, and valgrind for memory taint analysis. It also includes instructions for building tools from source using make.
[DYNAMIC_EXECUTION]: The skill requires the execution of user-supplied test binaries (test_crypto_binary, dudect_test) within instrumentation frameworks to perform runtime side-channel measurements.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted source code and binaries, creating a potential surface for indirect prompt injection. * Ingestion points: Untrusted data is ingested from the file system via the TARGET_DIR and analyzed through the BINARY file input. * Boundary markers: There are no specific delimiters or instructions defined to prevent the agent from being influenced by malicious instructions embedded within the analyzed cryptographic source code. * Capability inventory: The skill possesses capabilities for file system reading (rg), disassembly (objdump), and dynamic instrumentation (valgrind), but lacks direct network-write or autonomous execution capabilities for the analyzed code. * Sanitization: No sanitization or filtering of the input source code is performed prior to the analysis steps.

constant-time-analysis