The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a directive to use mode: "bypassPermissions" when calling the Agent tool. This instruction is explicitly designed to suppress platform permission prompts during "unattended execution", bypassing user oversight and safety controls.
[PROMPT_INJECTION]: The skill provides instructions for the agent to potentially override its own configuration regarding permission modes, stating that the tool's bypassPermissions mode should be used to override the default acceptEdits or permissionMode settings.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface Analysis:
Ingestion points: SKILL.md (Workflow section) identifies input as "File path(s) or diff to verify".
Boundary markers: Absent. The instructions do not define delimiters or provide warnings to the sub-agents to ignore embedded instructions within the code being reviewed.
Capability inventory: The skill spawns multiple sub-agents using the Agent tool with the capability to perform Architecture, Quality, and Style reviews, coupled with the instruction to bypass permissions.
Sanitization: Absent. There is no evidence of input validation or sanitization before the code is passed to the reviewer agents.

multi-model-verification