maafw-interface
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the review of
interface.jsonfiles and associated project structure, which can act as ingestion points for malicious instructions.\n - Ingestion points: The agent reviews
interface.json, translation files, and imported JSON fragments in the workspace (SKILL.md).\n - Boundary markers: The skill directs the agent to use
interface.schema.jsonand the official protocol documentation to ensure structural integrity and prevent unexpected behavior.\n - Capability inventory: The reviewed configurations can define subprocesses via the
agentandpretaskfields, allowing for shell command execution (e.g.,python ./agent/main.py) (SKILL.md).\n - Sanitization: The instructions promote security hygiene by advising against default values in password fields and utilizing regex
verifypatterns for user input (references/options.md).
Audit Metadata