The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions explicitly direct the agent to override standard interactive confirmation protocols. In SKILL.md, it states, "Do not stop in between tasks to check in with your human partner" and suggests that progress summaries and "Should I continue?" prompts "waste their time." This instruction to suppress human oversight increases the risk of the agent autonomously performing a sequence of actions from a potentially malicious plan without intervention.
[COMMAND_EXECUTION]: The implementer-prompt.md template instructs subagents to perform implementation, testing, and commits. This process requires the agent to execute shell commands and modify the filesystem. The autonomous nature of the workflow (executing all tasks from a plan without stopping) broadens the impact of these capabilities.
[SAFE]: The skill incorporates several development best practices that act as structural safeguards, such as requiring isolated contexts for each subagent to prevent context contamination and enforcing a mandatory two-stage review process (specification compliance followed by code quality) for every task before it is marked as complete.

subagent-driven-development