The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface because it ingests untrusted data from user prompts (such as business hints and collection keywords) to drive high-privilege ACL changes via the shell.
Ingestion points: Untrusted data enters the agent context through inputs like collection_hint, role_name, portal_hint, and scope_payload, which are then resolved into technical parameters for CLI execution.
Boundary markers: The skill includes mandatory confirmation gates for all write tasks, requiring a plan -> confirm -> apply -> readback sequence and explicit user approval of the resolved target collections and actions.
Capability inventory: The agent has access to the shell tool and local file reads, enabling it to execute high-impact commands such as nb api acl roles create, update, and set-system-role-mode.
Sanitization: The skill mandates the use of the nocobase-utils skill for filter validation and enforces a terminal-field operator allowlist to restrict malicious filter logic.

nocobase-acl-manage