The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a vendor-specific CLI tool (ae-cli) to perform administrative actions on the AE Agent system. These actions are governed by mandatory security rules, including role-based access checks, mandatory --dry-run usage to inspect effects before execution, and explicit user confirmation for all high-risk operations.
[EXTERNAL_DOWNLOADS]: The skill includes functionality to install software from the official npm registry via the +npm-install command. The skill enforces security boundaries by disabling npm lifecycle scripts by default (--ignore-scripts) and converting potentially dangerous symlinks to regular executable wrappers.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external integration sources (Feishu/Lark) and allows administrators to configure system prompts for communication channels, creating a surface for potential injection.
Ingestion points: Feishu identities resolved via OpenAPI MCP tools; batch rosters provided in bindings.json files.
Boundary markers: Implements a strict two-phase confirmation workflow (Configuration vs. Personnel binding) and mandatory dry-runs with credential redaction.
Capability inventory: Full system administration via ae-cli, including member management, sandbox lifecycle control, and remote tool installation.
Sanitization: Sensitive values and credentials are replaced with masking markers (***) in dry-run outputs; the underlying API enforces server-side validation of all configuration schemas.

ae-system