The Agent Skills Directory

[COMMAND_EXECUTION]: The bundled script scripts/cve_scan.py uses the subprocess module to execute native audit tools for various programming ecosystems (npm, pip, cargo, etc.). While these commands are standard for auditing, the --fix option triggers package update commands (e.g., composer update, go get, dart pub upgrade) which modify the local project environment and fetch external code as requested by the user.
Evidence: subprocess.run calls in scripts/cve_scan.py at line 133, using commands defined in the ECOSYSTEMS dictionary.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection because it ingests and displays external data (vulnerability titles and descriptions) originating from remote advisory databases.
Ingestion points: The script parses JSON output from external tools in parse_npm_audit (line 159), parse_pip_audit (line 192), and parse_cargo_audit (line 218).
Boundary markers: Findings are displayed in a formatted markdown-like table without explicit delimiters or instructions to the agent to ignore any commands embedded in the advisory text.
Capability inventory: The skill has the capability to execute shell commands through the subprocess module at line 133.
Sanitization: No sanitization is applied to the vulnerability descriptions or titles before they are presented in the report.

cve-scan