The Agent Skills Directory

[PERSISTENCE]: The skill utilizes the CronCreate tool to schedule recurring "Daily regression checks" that execute shell commands (npm test). This creates a mechanism where verification tasks and their associated commands continue to execute automatically in the background across independent sessions.
[INDIRECT_PROMPT_INJECTION]: The skill's core workflow relies on ingesting potentially untrusted external content, including project source code, git diffs, and browser screenshots. This content is then interpolated into prompts for multiple specialized sub-agents. This creates a significant attack surface for indirect prompt injection, where an attacker could hide malicious instructions in code comments, UI elements, or test data to influence agent grading, bypass quality gates, or manipulate the final merge verdict.
[COMMAND_EXECUTION]: As part of its verification logic, the skill executes arbitrary shell commands defined within the project being verified (e.g., test runners like npm test or pytest, and linters like ruff or biome). While this is central to the skill's purpose, it allows for the execution of arbitrary code found within the project environment.
[DATA_EXFILTRATION]: The skill uses the PushNotification tool to send status updates and verification summaries to external notification services. Additionally, it records verification results and metrics in the memory MCP server, which may include sensitive information about the project structure or codebase history.

verify