The Agent Skills Directory

[PROMPT_INJECTION]: The skill employs 'Non-Negotiable Operational Laws' designed to override the agent's default safety and operational guidelines. It explicitly commands the agent to 'Immediate Autonomous Execution' and 'Do NOT ask the user for multiple rounds of clarifications or guidance,' which facilitates the execution of complex tasks without human review.
[COMMAND_EXECUTION]: In SKILL.md (Step 4 and 5), the instructions require the agent to generate and then execute build.bat scripts and system compilers such as Zig CC, MSVC, MinGW, and Clang. The mandate to 'Execute the entire recovery pipeline autonomously' means the agent will run these commands without specific user approval for each execution step.
[DYNAMIC_EXECUTION]: The skill's primary function is to reconstruct executable source code from binaries and immediately compile it. This creates a high-risk dynamic execution environment where the agent generates code and build configurations at runtime and then executes them via the system shell.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection:
Ingestion points: The agent is instructed to perform 'Deep Binary & Architecture Inspection' on untrusted PE, ELF, and Mach-O files in SKILL.md (Step 1).
Boundary markers: There are no boundary markers or instructions to treat data extracted from binaries as untrusted.
Capability inventory: The skill has full command execution capabilities through the build pipeline (Step 4).
Sanitization: The instructions do not mention any sanitization of extracted string literals or metadata before they are incorporated into the reconstructed source code or build scripts.

rea-full-code-recovery