The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documentation introduces specialized tools like fxmind_fivem_nui_dump that allow an agent to ingest the structured state of the UI for analysis. This creates a surface for indirect prompt injection where malicious instructions embedded in game data (such as player names or custom item descriptions) could influence the agent's behavior during debugging sessions.
Ingestion points: fxmind_fivem_nui_dump and registerFxmindNuiDump (in SKILL.md) allow the agent to read the Zustand store and DOM state.
Boundary markers: Absent; there are no specific instructions for the agent to treat the collected NUI state as untrusted data or to use delimiters.
Capability inventory: The skill describes tools like fxmind_fivem_nui_wire (which modifies project files) and uses Post.create to send data back to the game environment.
Sanitization: Absent; the provided code snippets do not include sanitization or escaping of the data extracted from the UI.
[COMMAND_EXECUTION]: The skill describes an automated "wiring" process using the fxmind_fivem_nui_wire tool, which is intended to patch the fxmanifest.lua file and inject a DOM probe into the project's source code. While this is a development-oriented feature, automated modification of project files by an agent should be monitored to ensure it remains within the intended scope.

fivem-react-nui