The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process external data (requirements, design documents, acceptance criteria) to generate implementation plans for subagents. While these plans are intended for software development, there is an inherent surface for indirect prompt injection if the source requirements contain malicious instructions that the agent reproduces in the task plan. The instructions lack explicit boundary markers or sanitization steps to filter input from potentially untrusted requirement documents before interpolation into the output plan.
[COMMAND_EXECUTION]: The skill provides templates and examples involving shell commands such as git, pytest, and mkdir. These are documented as standard development workflows for task implementation and verification. The agent is encouraged to use these tools to explore the codebase and commit changes, representing the standard capability tier for developer-oriented agents.

writing-plans