The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates entirely by executing the 'bl' CLI binary to interact with Alibaba Cloud Model Studio services.
Evidence: The documentation across reference files (e.g., reference/app.md, reference/text.md, reference/knowledge.md) defines extensive patterns for the AI agent to execute 'bl' commands for resource management and model interactions.
[EXTERNAL_DOWNLOADS]: The skill provides native functionality to install additional Command Packs and extensions from official registries.
Evidence: Commands such as bl plugin install (in reference/plugin.md) and bl skill add (in reference/skill.md) facilitate the downloading of external extensions from the Bailian skill registry and scoped packages like '@ali/'.
[CREDENTIALS_UNSAFE]: The skill manages sensitive authentication data including API keys and Alibaba Cloud AccessKey ID/Secrets for cloud environment access.
Evidence: The reference documentation for 'bl auth login' and 'bl config set' (in reference/auth.md and reference/config.md) provides instructions for the agent to store and configure these credentials for CLI usage.
[DYNAMIC_EXECUTION]: The skill allows for the execution of dynamic workflow definitions and arbitrary console API calls.
Evidence: The bl pipeline run command (in reference/pipeline.md) executes JSON/YAML pipeline definitions, and bl console call (in reference/console.md) enables dynamic interaction with the cloud console API gateway.
[INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection by ingesting and processing user-controlled text prompts for model execution.
Ingestion points: User prompts and messages passed to bl app call and bl text chat in SKILL.md and its reference files.
Boundary markers: The skill does not explicitly define delimiters or instruction-ignore wrappers for user-provided data within the CLI flags.
Capability inventory: The skill possesses extensive system capabilities via the 'bl' CLI, including network communications, configuration modifications, and credential handling.
Sanitization: Sanitization is handled by the underlying 'bl' CLI tool and the Alibaba Cloud Model Studio platform guardrails.

bailian-cli