The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or unauthorized data access were detected. The skill is well-documented and implements intended functionality for the Personize ecosystem.
[EXTERNAL_DOWNLOADS]: The skill interacts with https://agent.personize.ai for data storage, retrieval, and governance. This domain is the official API endpoint for the skill author and is treated as a trusted resource.
[PROMPT_INJECTION]: The skill implements a multi-agent collaboration surface (Category 8: Indirect Prompt Injection) by design, where agents process data contributed to a shared workspace. This is the primary function of the skill and is managed through the use of structured JSON entries and the 'Three-Layer Operating Model' (Guidelines + Memory + Workspace).
Ingestion points: Data is ingested from the shared workspace via client.memory.smartDigest and client.memory.smartRecall in recipes/multi-agent-account.ts and recipes/trigger-dev-bridge.ts.
Boundary markers: Boundary markers are not explicitly defined in the example prompts, but the protocol requires agents to call ai_smart_guidelines to retrieve organizational constraints before processing workspace data.
Capability inventory: Agents possess the capability to write to the data store via memorize and generate actions using the prompt API.
Sanitization: The skill uses JSON.stringify for storage and a custom safeParseJSON helper in recipes/helpers.ts to validate and clean LLM-generated JSON before processing.

personize-agent-workspace