The Agent Skills Directory

[DYNAMIC_EXECUTION]: The methodology requires the agent to modify source code and verify changes by running local build tools like make, CMake, or cargo (found in references/source-modify.md). Running build processes on modified code is a form of dynamic execution that can be risky.
[INDIRECT_PROMPT_INJECTION]: The skill fetches and analyzes external security rules from public repositories, creating a surface for indirect instructions. Ingestion points: Security rules are retrieved via gh search and http_request (references/detection-search.md). Boundary markers: There are no mechanisms to isolate the external rule content or prevent the agent from following instructions potentially hidden in rule comments. Capability inventory: The agent has capabilities to read/write files and execute shell commands/build scripts. Sanitization: No input sanitization or validation is applied to the retrieved rules before they are parsed.
[COMMAND_EXECUTION]: The skill uses shell commands like find, grep, xxd, and objdump for analyzing local binaries and source code. These commands could be vulnerable to manipulation if search patterns are derived from untrusted input.
[EXTERNAL_DOWNLOADS]: The skill provides commands to search for and download security rules from third-party GitHub organizations. While targeting specific repositories, the content is not controlled by the skill author.

c2-evasion-methodology