The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of the rozenite CLI tool to perform all operations. Evidence: SKILL.md and various domain files (e.g., console.md, network.md) define workflows using rozenite agent <domain> call and other CLI commands.
[DATA_EXFILTRATION]: The skill provides capabilities to read potentially sensitive data from the target application's environment, including filesystem access and storage inspection. Evidence: domains/file-system.md includes tools like read-text-file and read-image-file which accept arbitrary paths; domains/storage.md and domains/mmkv.md allow reading storage entries.
[PROMPT_INJECTION]: The skill processes untrusted data from multiple sources, creating an attack surface for indirect prompt injection.
Ingestion points: Untrusted data is ingested via console logs (domains/console.md), network traffic recordings (domains/network.md), and application storage/files (domains/storage.md, domains/file-system.md).
Boundary markers: No explicit boundary markers or instructions to disregard embedded instructions are defined in the prompt instructions.
Capability inventory: The agent can execute domain-specific tools with JSON arguments using the rozenite agent <domain> call command, allowing it to mutate application state or storage.
Sanitization: There is no evidence of sanitization or validation of the retrieved data before it is processed or used by the agent.

rozenite-agent