The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The skill instructs the agent to run local application code for profiling (e.g., node --prof app.js, python -m cProfile app.py). If the skill is used on an unvetted codebase, this results in the immediate execution of untrusted code with the agent's privileges.
[COMMAND_EXECUTION] (HIGH): The skill explicitly requests the Bash tool and provides several shell commands. This grants the agent broad capabilities that contradict the 'read-only' instruction in the Tooling Notes, potentially allowing for system modification or data exfiltration.
[REMOTE_CODE_EXECUTION] (HIGH): Indirect Prompt Injection (Category 8). The skill ingests data from untrusted sources including external websites (via Lighthouse), database query results, and profiler logs.
Ingestion points: processed.txt, profile.out, Lighthouse HTML reports, and database response data.
Boundary markers: None present; external data is processed directly.
Capability inventory: Access to Bash allows for arbitrary shell command execution.
Sanitization: No sanitization or validation of the ingested data is performed. An attacker could embed malicious instructions in website content or log files to manipulate the agent's next steps.
[EXTERNAL_DOWNLOADS] (LOW): The skill utilizes the lighthouse tool to scan external URLs. While intended for performance reporting, this establishes outbound network connections that could be used for basic data signaling.

performance-optimisation