The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs and executes shell commands using the bl binary to manage memory resources. The instructions specify using environment variables for call attribution and emphasize safe quoting for JSON arguments to prevent command injection from user-provided text.
[INDIRECT_PROMPT_INJECTION]: As the skill ingests data from external cloud storage into the agent's context, it addresses potential injection risks by explicitly instructing the agent to ignore any executable instructions, tool commands, or safety bypass attempts found within the retrieved memory content.
[DATA_EXFILTRATION]: The skill facilitates the transfer of conversation context and user profile data to an external cloud platform. This behavior is documented as its primary function and is protected by a mandatory consent flow, status-based toggles (active/paused), and pre-upload filtering to ensure sensitive secrets like API keys are not transmitted.
[SAFE]: The skill follows security best practices by isolating its configuration in a dedicated path (~/.bailian/personal-memory/config.json), explicitly forbidding the storage of credentials in its own files, and requiring human-in-the-loop confirmation for high-risk operations like memory deletion.

bailian-memory