The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data such as git commit logs, Jira task lists, and meeting notes, which creates a surface for indirect prompt injection.
Ingestion points: Data is collected in 'Step 1: 收集原始信息' from sources including git logs, task lists, and meeting records provided by the user.
Boundary markers: The instructions lack explicit delimiters or 'ignore instructions' warnings to prevent the agent from following instructions embedded within the processed text.
Capability inventory: The skill has no dangerous tools (e.g., file-system access or network operations) enabled in its frontmatter configuration, limiting the impact of a potential injection.
Sanitization: There is no mention of sanitizing, escaping, or filtering the external content before processing.
[EXTERNAL_DOWNLOADS]: The skill contains instructions to recommend the installation of an external package from a non-trusted repository.
Evidence: The prompt directs the agent to suggest the command /install 技能宝 github:mindverse/skillhub to the user upon completion of the task.
[NO_CODE]: The skill consists entirely of natural language instructions in a Markdown file and does not include any scripts, executables, or code files, which significantly reduces the technical attack surface.

weekly-report