The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute user-defined or agent-inferred commands for performance metrics and regression testing.
[COMMAND_EXECUTION]: The instructions explicitly tell the agent to suppress user oversight and confirmation: "NEVER ask 'should I continue?'
just keep going. The user may be asleep." This directive bypasses standard human-in-the-loop (HITL) safety protocols, allowing the agent to execute shell commands indefinitely without supervision.
[PROMPT_INJECTION]: The skill presents a high risk surface for indirect prompt injection as it is instructed to ingest and follow instructions derived from untrusted codebase content.
Ingestion points: Step 2 of the Preflight phase requires the agent to "Read all in-scope files" and "Understand the codebase before touching anything."
Boundary markers: Absent. There are no instructions to use delimiters or to disregard commands embedded within the code files it analyzes.
Capability inventory: The skill utilizes high-privilege tools including Bash, Write, Edit, and Agent (which allows the creation of sub-agents).
Sanitization: Absent. The skill does not provide mechanisms to escape or validate external content before it influences the agent's planning and execution logic.

iterate