grind
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_CONTEXT_INJECTION
Full Analysis
- [DYNAMIC_CONTEXT_INJECTION]: The
SKILL.mdfile utilizes Claude Code's dynamic context syntax (!\command`) to executeskill-toc`. This is used legitimately for generating a table of contents for the skill's scripts and references at load time. - [COMMAND_EXECUTION]: The skill's primary function is to execute shell commands (provided via
--check,--gate, and--runnerarguments) in a loop. These commands are user-provided at launch and are essential to the skill's intended purpose of automating long-running workflows. - [PRIVILEGE_ESCALATION]: The skill explicitly defines a boundary between the agent and the operator. It prevents the agent from stopping the loop via the
stopsubcommand by checking for an environment variable (GRIND_ITERATION) and refusing execution if it's present, ensuring that only a human operator can terminate the process through standard channels. - [DATA_EXPOSURE]: The skill uses an append-only journal to maintain state. While it accesses this file frequently, it does not interact with sensitive system directories (like
.sshor.aws) or hardcode credentials.
Audit Metadata