The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions to use mode: "bypassPermissions" to override safety and permission filters. It specifically references a 'Universal bypassPermissions' pattern (R010) intended to circumvent platform constraints.
[PRIVILEGE_ESCALATION]: The instructions direct the agent to pass a specific bypass flag during sub-agent calls to avoid user confirmation (acceptEdits). This represents an attempt to escalate capabilities beyond the default restricted environment and enable unattended execution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input via the component description and interpolates it into prompts for multiple parallel agents without sanitization.
Ingestion points: The <component/page description> argument defined in the frontmatter argument-hint.
Boundary markers: None. The input is passed directly to sub-agents in Phase 2.
Capability inventory: Spawns multiple parallel agents (R009) to generate and display code.
Sanitization: No escaping or validation is performed on the user-provided description before it is processed by the generation agents.

design-shotgun