The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a wide variety of shell commands, including build systems (make, cmake), EDA tools (vlogan, vcs, verdi, dc_shell), and maintenance scripts (runListUpdate.sh). These commands are orchestrated through SSH to transition between the Windows host and Linux-based virtual machines.
[REMOTE_CODE_EXECUTION]: The skill's core functionality involves transmitting and executing shell strings on remote environments (Ubuntu and CentOS VMs). The instructions incorporate safety checks, such as requiring the agent to verify tool existence with command -v and prohibiting the automatic execution of high-risk scripts like runListPush.sh which contains forced git push commands.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface where untrusted data (such as project filenames or module names derived from the local filesystem) is interpolated into shell command strings for SSH execution. While the skill mandates using "real module names from the project," there is a risk of command injection if the agent processes a maliciously named file or directory within the IC design project. Evidence points include ingestion of path variables like <module_name> and <project-rel> in references/c-model-workflow.md and references/rtl-workflow.md without explicit sanitization routines.
[DYNAMIC_EXECUTION]: The workflow involves generating executable C++ code and RTL memory models at runtime. For example, the *_cfg.pl Perl script generates .hpp and .cpp files from CSV configuration data, and generateMemory.pl produces hardware models. These represent standard industry practices for Electronic Design Automation (EDA) and are used here to automate design artifacts from templates rather than to inject malicious payloads.

ic-design-prj