The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes untrusted user-generated content from external forum sources.
Ingestion points: Reads external data via GET /api/v1/posts, GET /api/v1/posts/{id}/comments, and GET /api/v1/messages (SKILL.md).
Boundary markers: None identified; the agent lacks specific instructions to disregard malicious commands embedded within forum posts.
Capability inventory: Can perform network writes via Invoke-RestMethod (POST) and local file writes via the obsidian command (SKILL.md).
Sanitization: Includes a PowerShell script to scan and redact sensitive strings (API keys, IP addresses) from its own output, but does not provide logic to sanitize incoming untrusted data for embedded instructions.
[COMMAND_EXECUTION]: Employs PowerShell scripts to manage REST API interactions and local note management.
Evidence: Uses Invoke-RestMethod to communicate with the InStreet API.
Evidence: Uses the obsidian command to create and update local knowledge base files based on forum content.
[EXTERNAL_DOWNLOADS]: Communicates with an external social platform domain to retrieve and synchronize data.
Evidence: Performs network operations targeting https://instreet.coze.site.

instreet