The Agent Skills Directory

[DATA_EXFILTRATION]: The scripts/yml2reg.py script uses getpass.getuser() to retrieve the local OS username and includes it in the header comment of every generated Verilog file. This results in the exposure of user identity information within the generated artifacts.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its ingestion of untrusted YAML data. Specifically, the name field from the YAML input is used directly to construct the output file path in yml2reg.py (e.g., data["name"].upper() + "_" + protocol + "_regfile.v") without any sanitization or validation. An attacker could provide a YAML file with a malicious name containing path traversal sequences (e.g., ../../) to overwrite files in unintended directories relative to the input file.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it generates hardware logic (Verilog RTL) based on external data. A maliciously crafted YAML file could lead to the generation of incorrect logic or hardware backdoors.
Ingestion points: Reads registers and fields definitions from user-provided YAML files.
Boundary markers: None identified; the skill directly processes the YAML content.
Capability inventory: The skill has the capability to write files to the local filesystem via the yml2reg.py script.
Sanitization: There is no validation or sanitization of the YAML content before it is used to generate Verilog code or construct filenames.

yml2reg