architect-update-config-safely
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes
curlcommands to interact with the ElevenLabs API atapi.elevenlabs.io. These commands are used to retrieve current agent settings and perform partial updates using thePATCHmethod. Authentication is implemented using a variable placeholder for the API key.\n- [INDIRECT_PROMPT_INJECTION]: The skill identifies a data ingestion surface by reading existing configuration state from the API to inform future updates. Findings for this category are considered safe given the trusted source and the skill's primary purpose.\n - Ingestion points: Data enters the context from the ElevenLabs API via
GET /v1/convai/agentsinSKILL.md.\n - Boundary markers: Absent.\n
- Capability inventory: Includes network operations via
curlto perform configuration patches on the ElevenLabs API inSKILL.md.\n - Sanitization: The instructions require the agent to confirm exact key names and native data types from the current config response before applying changes.
Audit Metadata