The Agent Skills Directory

[SAFE]: The skill uses a secure authentication model where the 博客园 personal access token is injected via a connector as the $CNBLOGS_TOKEN environment variable. The code follows the official cnblogs/vscode-cnb client contract and explicitly instructs the agent never to print the token.
[SAFE]: All write and delete operations (creating, updating, or deleting posts) are protected by a mandatory confirmation flow. The CLI defaults to a dry-run mode, requiring the --confirm flag as the final argument to execute the actual API request. This ensures the user reviews all content and settings before publication.
[SAFE]: The implementation of CNBlogsClient includes a custom NoRedirectHandler. This prevents the Authorization header, which contains the user's PAT, from being automatically forwarded to external domains in the event of an HTTP redirect, mitigating potential credential leakage.
[INDIRECT_PROMPT_INJECTION]: The skill ingests Markdown content from external files or user-provided strings for publication. While this constitutes an indirect prompt injection surface where embedded instructions could target the agent, the risk is effectively mitigated by the skill's requirement for manual user confirmation and full content preview before any data is sent to the CNBlogs API.

cnblogs