The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from PRDs, alert groups, and knowledge bases. While it lacks explicit boundary markers for these data sources, it mitigates risk by instructing the agent to avoid jumping to conclusions and strictly limiting its role to generating suggestions and scripts for human review.
[COMMAND_EXECUTION]: The skill is designed to generate data correction scripts (SQL) and unit tests. It includes robust security guidelines, such as the 'Select before Update' hard rule and the requirement that all executions must be performed manually by the user, preventing unauthorized or automated production changes.
[DATA_EXFILTRATION]: The skill aggregates information from developer tools (Lark, Gitee, Linear). The analysis found no evidence of exfiltration patterns; the skill directs output to user-specified document locations within the authorized workspace environment.

job-dev