brida-reflex-game-state-action
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions recommend checking the vendor's live status endpoints, such as
/reflex/llms.txtand/reflex/agent.json. These resources are provided by the skill author, brida-ai, to ensure the agent has access to current service availability and contract versions.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a potential surface for indirect prompt injection as it processes dynamic game state data from the user's project environment.\n - Ingestion points: Untrusted game state data enters the system as the primary input for tactical decision-making in
SKILL.md.\n - Boundary markers: The logic is strictly constrained by a JSON contract in
references/custom-reflex.jsonthat limits the AI's choices to a specific, bounded action set (advance,defend,retreat,hold).\n - Capability inventory: The skill utilizes platform tools (MCP, SDKs, or REST) to perform state inspection and execute validation fixtures.\n
- Sanitization: The skill includes a clear authority boundary, stating that the agent's output is semantic evidence only and must be revalidated by the host game engine, preventing the AI from executing arbitrary code or unauthorized actions.
Audit Metadata