The Agent Skills Directory

[COMMAND_EXECUTION]: The validation script scripts/agent.py uses subprocess.run to execute PowerShell commands for verifying certificate deployment. The internal CA Common Name (internal_ca_cn) is interpolated directly into the command string using an f-string without any sanitization or escaping. This allows for arbitrary command injection if an attacker can influence the CA name argument passed to the script.
Evidence: scripts/agent.py lines 97-101: subprocess.run(["powershell", ... f'Where-Object {{$_.Subject -like "*{self.internal_ca_cn}*"}}' ...]).
[PRIVILEGE_ESCALATION]: The instructions for configuring macOS and Linux endpoints in SKILL.md require the execution of commands with sudo to add certificates to the system-wide trust store. While this is a standard requirement for the described task, it involves the agent performing high-privilege operations that grant administrative control over the local machine's security configuration.
Evidence: SKILL.md Step 2: sudo security add-trusted-cert and sudo cp ssl-inspect-ca.crt /usr/local/share/ca-certificates/.
[INDIRECT_PROMPT_INJECTION]: The agent.py script provides a vulnerability surface by ingesting hostnames and CA names from command-line arguments and using them in network connections and shell command assembly. The script lacks input validation, creating a risk if the agent is directed to process untrusted data retrieved from external sources before calling the validation tool.
Ingestion points: sys.argv in main() function of scripts/agent.py.
Boundary markers: None present.
Capability inventory: subprocess.run (shell execution) and ssl.create_default_context (network connectivity) in scripts/agent.py.
Sanitization: No input validation or escaping is performed on external arguments.

performing-ssl-tls-inspection-configuration