The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute harness scripts and interact with a live instance of the application during the 'Live pass' phase. This behavior is fundamental to the skill's verification purpose but involves running arbitrary code found within the target project.
[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting and analyzing untrusted external data, specifically project source code and feature documentation files. This creates an attack surface where maliciously crafted comments or instructions within those files could attempt to subvert the agent's logic during the audit.
Ingestion points: Project source files, feature map READMEs, and harness scripts are read and processed in the 'Source wave' and 'Index hygiene' steps.
Boundary markers: The skill does not define specific delimiters or instructions to prevent the agent from obeying potential directives embedded within the analyzed project files.
Capability inventory: The agent possesses the capability to launch subagents, execute local shell commands (via harnesses), and submit code changes through Pull Requests.
Sanitization: There is no evidence of sanitization or filtering applied to the content extracted from the project files before it is processed by the agent.

maintain-verification-skill