The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests arbitrary data from user conversations and technical plans to generate Markdown-based task specifications for subsequent coder agents. This process creates an attack surface where malicious or deceptive instructions in the source conversation could be reformatted and 'promoted' into the final specification files. Since these files are designed to be used independently by agents to implement features, such poisoned tasks could lead to the introduction of security vulnerabilities, malicious dependencies, or backdoors in the resulting software without immediate detection.

create-spec