The Agent Skills Directory

Indirect Prompt Injection Surface: The skill allows the agent to drive VS Code Chat using user-provided prompt templates (via the --message argument in chat-memory-smoke.mts). Since the agent can invoke tools like terminal commands within the VS Code environment, there is a potential surface for indirect prompt injection if malicious instructions are embedded in the data processed during a run. The skill documentation correctly addresses this risk by advising the use of throwaway workspaces and safe, read-only commands.\n- Handling of Sensitive User Profiles: The utility in userDataProfile.mts facilitates the copying of VS Code user-data profiles, which typically contain authentication secrets such as Copilot tokens. Although this is intended to streamline performance testing by reusing logged-in sessions, it introduces a risk of credential exposure. The skill includes warnings to keep these profiles within local, git-ignored directories and to never share them.\n- Heap Snapshot Data Exposure: The automation scripts capture renderer heap snapshots to analyze memory leaks. These snapshots are comprehensive memory dumps that may contain sensitive strings, session data, or code fragments present in the process memory at the time of capture. Users should be aware that sharing these artifacts could lead to unintended data exposure.\n- Local Command and Tool Execution: The skill requires the ability to launch VS Code processes and execute local scripts and tools (e.g., npx @playwright/cli). This involves high-privilege operations on the user's machine, such as enabling the smoke-test driver and remote debugging ports, which should only be performed in a trusted environment.

auto-perf-optimize