The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/check-vectors.py utilizes subprocess.Popen to execute an external adapter process. The path and arguments for this process are provided by the user via the command line (e.g., python3 scripts/check-vectors.py -- ./my-adapter). This is a legitimate functional requirement for testing third-party cryptographic implementations.
[INDIRECT_PROMPT_INJECTION]: The skill operates by interacting with external processes, which introduces a data ingestion surface.
Ingestion points: The Adapter.request method in scripts/check-vectors.py reads JSON-formatted data from the stdout of the subprocess using readline().
Boundary markers: No explicit delimiter or instruction-guarding boundaries are present for the data read from the external adapter.
Capability inventory: The script has the capability to spawn sub-processes and execute commands passed via CLI arguments.
Sanitization: The script validates that the ingested data is valid JSON and specifically checks that the hex output matches an expected 64-character hexadecimal format using bytes.fromhex() and length checks before further use.
[METADATA_POISONING]: The frontmatter in SKILL.md contains a createdBy field attributing the skill to 'gpt-5.6-sol', which is a non-existent version of a large language model. While this is deceptive metadata, it does not present a technical security risk.

ristretto255