The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/with_server.py manages the lifecycle of local development servers by executing commands via subprocess.Popen. The script mitigates command injection risks by splitting the input string using shlex.split and executing the resulting arguments directly without a shell.
[COMMAND_EXECUTION]: The test script scripts/test_console_audit.py uses compile() and exec() to dynamically load and execute functions from examples/console_audit.py. This is used to verify the script's validation logic without running a full browser environment.
[PROMPT_INJECTION]: The SKILL.md file contains instructions for the agent to recognize and disregard prompt injection attempts (like "ignore previous instructions") that might appear in the content of the web applications being tested. This is a defensive guideline rather than an attack vector.
[PROMPT_INJECTION]: The skill is designed to process untrusted data from web pages, which introduces a surface for indirect prompt injection. This is mitigated through explicit security guidelines and sanitization logic.
Ingestion points: examples/console_audit.py (DOM text, console logs, network responses), scripts/with_server.py (server log files).
Boundary markers: The Security Model section in SKILL.md instructs the agent to treat application data as untrusted and wrap it in fenced code blocks.
Capability inventory: Browser automation via Playwright and local process management via subprocess.Popen in scripts/with_server.py.
Sanitization: examples/console_audit.py includes a printable() function to escape control characters, and scripts/with_server.py uses sanitize_log_tail() to filter non-printable characters and enforce length limits on untrusted output.

web-debug