The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to read and analyze CPU profile logs (.json and .md summary files) generated from a web application. These logs contain function names and metadata derived directly from the code of the website being profiled. An attacker-controlled website could include malicious strings in its source code that appear in the profiling logs, attempting to influence the agent's behavior during the analysis phase.
Ingestion points: The agent is instructed to read generated files in ~/.cursor/browser-logs/ (Step 6).
Boundary markers: There are no explicit instructions or delimiters used to warn the agent to ignore embedded instructions within the profile logs.
Capability inventory: The skill utilizes the cursor-ide-browser toolset, including browser_navigate, browser_click, browser_scroll, and browser_fill (Steps 2 and 4).
Sanitization: No sanitization or validation of the profiling data is performed before it is processed by the agent.

profiling-performance