The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The assets/ci-config-template.yml file contains a remote script execution pattern where an installer is downloaded and piped directly to bash: curl -s https://raw.githubusercontent.com/aquasecurity/tfsec/master/scripts/install_linux.sh | bash. While targeting a well-known security vendor's repository, the use of a mutable branch reference combined with direct shell execution is a high-risk practice.
[COMMAND_EXECUTION]: The skill recommends binding the mitmproxy and mitmweb services to all network interfaces using --listen-host 0.0.0.0 in several workflow examples. This configuration allows any host on the same network to potentially access the proxy and intercept or modify traffic without authentication.
[COMMAND_EXECUTION]: In the Troubleshooting section of SKILL.md, the skill recommends the command rm -rf ~/.mitmproxy/ to resolve SSL certificate errors. This is a destructive operation that recursively deletes the entire application configuration and certificate directory without any safety warnings or confirmation prompts.
[DATA_EXFILTRATION]: The Python script example AuthTester in SKILL.md captures and prints the first 20 characters of authentication headers. This results in sensitive credential material being logged to the console or CI output, potentially leading to session token exposure.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (API traffic) and provides the agent with capabilities to modify and replay this traffic. There is a lack of boundary markers or sanitization logic in the provided script templates to prevent an attacker-controlled API response from influencing the agent's subsequent analysis or actions.
Ingestion points: Intercepted HTTP/HTTPS flows and WebSocket messages processed via Python addons (e.g., api-test.py, auth-test.py).
Boundary markers: Absent. The skill does not instruct the agent to ignore or delimit instructions found within captured traffic.
Capability inventory: The skill uses mitmproxy which has full capabilities to read, write, and replay network traffic, and uses Python addons that can execute arbitrary logic on flow objects.
Sanitization: Absent. Data from flow.request.text or flow.response.text is processed without filtering.

api-mitmproxy