The Agent Skills Directory

[SAFE]: The skill sets proactive security constraints for the agent, such as forbidding the output of authentication secrets and requiring explicit user intent for data modification or deletion. It follows best practices for secret management by instructing the agent to never output the old API key and to use official CLI commands for initialization.\n- [INDIRECT_PROMPT_INJECTION]: The skill defines rules for processing external tool outputs. It includes instructions for verifying response codes and handling error states, which reduces the surface for indirect injection. Evidence of ingestion points can be found in SKILL.md under '响应解析规则', where the agent is told to check for 'code', 'status', and 'reason' before parsing data. Capability inventory includes subprocess calls via 'mocli'. Sanitization is handled by the strict requirement to verify return codes and not fabricate results if data is missing.\n- [EXTERNAL_DOWNLOADS]: The reference material 'references/mocli-output-proto.md' mentions an update command for the vendor's package ('@mowenxd/cli'). This is a legitimate vendor resource. The instructions in SKILL.md under '更新提醒' specifically restrict the agent from running this automatically, maintaining user control and preventing unauthorized execution of update scripts.

mo-shared