The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the bl (Bailian CLI) binary to perform all resource management tasks, including applying infrastructure changes, destroying resources, and managing project builds.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading skill versions and project files from the Alibaba Cloud Bailian service through commands like bl managed-agent skill download and bl managed-agent file download. These actions are part of the standard managed-service workflow.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a potential attack surface for indirect prompt injection as it ingests and processes untrusted local files (such as agents.yaml and SKILL.md) and remote session event logs.
Ingestion points: Data enters the agent's context during the parsing of infrastructure declarations in agents.yaml, custom skill files, and when retrieving message history using bl managed-agent session event list.
Boundary markers: The skill instructions emphasize critical safety guardrails, requiring the agent to show plans or diffs to the user and wait for explicit manual confirmation before executing any command with the --yes flag.
Capability inventory: The tool set includes the ability to write project files locally, communicate with cloud APIs for resource deployment, and execute shell commands via the bl binary.
Sanitization: The instructions explicitly direct the agent to redact secrets from diagnostic exports and previews, and use environment variable references instead of hardcoded values to minimize the risk of credential exposure.

bailian-managed-agent