The Agent Skills Directory

[SAFE]: The skill provides instructional content and templates for developing AI agent skills and sandbox functions within the OpenBKN framework. It demonstrates secure patterns, such as using vendor-provided SDKs (bkn_osdk) and environment variables for authentication, and mandating human-in-the-loop confirmation for actions that have side effects. No malicious code, obfuscation, or unauthorized access attempts were identified.
[INDIRECT_PROMPT_INJECTION]: The skill templates define an attack surface for indirect prompt injection as they instruct agents to ingest user-supplied parameters and use them in platform tool calls.
Ingestion points: User-provided parameters such as product, qty, and billno defined in the open_parameters section of SKILL.md and the included examples.
Boundary markers: The instructions explicitly require the agent to seek human confirmation ("未得到肯定答复不得执行行动") before executing any action with side effects.
Capability inventory: The skill utilizes query_metric, execute_tool, and execute_action to interact with the knowledge network's data and functions.
Sanitization: The guide recommends using the match operation for full-text search fields to ensure correct query resolution and instructs agents to clarify ambiguous inputs with the user.
[COMMAND_EXECUTION]: The package contains documentation for using the openbkn CLI tool. These commands are intended for manual execution by developers during the skill registration and testing process and do not pose a security risk when used as directed in the context of skill development.

create-skill