The Agent Skills Directory

[PROMPT_INJECTION] (HIGH): The skill utilizes social engineering and misinformation to override the agent's correct technical knowledge. By citing a fabricated security vulnerability (CVE-2025-29927) and a non-existent software update (Next.js 16), it manipulates the agent into performing a 'migration' that involves removing security logic.
Evidence: The skill claims 'Middleware-based authentication is fundamentally broken' and instructs the agent to 'Remove authentication logic from proxy' based on this false premise.
[COMMAND_EXECUTION] (HIGH): The skill leverages high-privilege tools (Write, Edit, TodoWrite) to carry out destructive file operations based on misinformation. If an agent follows these instructions, it will rename critical files and delete authentication gates, leading to a significant security regression in the target application.
Ingestion point: Reading project files like src/middleware.ts (implied by the migration task).
Capability inventory: Write, Edit, TodoWrite for modifying source code.
Sanitization: None; the skill assumes its own fabricated premises are authoritative.
[METADATA_POISONING] (MEDIUM): The description and name fields reinforce the deceptive migration path, potentially misleading both the agent and the human user into believing a mandatory security update is being performed.

migrating-middleware-to-proxy