The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze external data sources such as contract documents, Git history, and architectural decisions. While the skill lacks explicit boundary markers or sanitization instructions for these inputs, it contains no dangerous capabilities (e.g., file system modifications or arbitrary code execution) that could be exploited via malicious content in those documents. The risk is limited to potential bias in the generated legal matrix or memo, which is mitigated by the skill's requirement to verify all citations against official legal databases like gesetze-im-internet.de and dejure.org.

miturheber-mehrere-model-weights-nearshore