The Agent Skills Directory

[COMMAND_EXECUTION]: Extensively utilizes the bl CLI tool to perform cloud-based operations including text generation, model management, and workspace configuration across multiple files in the reference/ directory.- [EXTERNAL_DOWNLOADS]: The bl skill add and bl plugin install commands facilitate the download and installation of additional skills and command packs from the official vendor registry (Bailian registry).- [DATA_EXFILTRATION]: Provides functionality to read local files (images, video, CSV, JSONL) and upload them to Alibaba Cloud (DashScope/Bailian) infrastructure through commands like bl file upload and bl batch create.- [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect injection by ingesting and processing untrusted data from user-provided files and prompts.
Ingestion points: Input files provided to bl batch convert (--file), bl decision (--state-file, --questions-file), and bl app call (--prompt).
Boundary markers: No explicit boundary markers or 'ignore' instructions are defined in the skill's markdown; safety relies on the CLI's internal handling.
Capability inventory: Wide-ranging shell command execution for cloud resource management and model invocation.
Sanitization: No specific input sanitization or validation logic is described in the provided instructions.- [PERSISTENCE]: The bl mcp connect command modifies the native configuration files of other local AI agents (such as Claude Code, Cursor, and QwenCode) to register MCP servers, which alters the persistent behavior and capabilities of those tools.

bailian-cli