/goal (Claude Code)
The Claude Code command that keeps the agent working until a condition holds, without you sending a message at every step. You write the condition ("all tests in test/auth pass and lint is clean"), Claude chains turns, and after each one a small fast model, Haiku by default, reads the conversation and returns a verdict: not yet, met, or impossible. The agent does not decide it is done, a second model does. `/goal` with no argument shows the status, `/goal clear` stops it.
Strengths
- The finish is judged by a model separate from the one doing the work
- Nothing to install: one command, active for the current session
- Verdict and reason visible every turn, with turn count and token spend
Limitations
- The evaluator only sees the conversation: a condition Claude never surfaces is never validated
- In manual mode Claude still asks before each tool, you need auto mode to run unattended
Best for
- Migrating a module until every call site compiles and tests pass
- Working through a labelled ticket queue until it is empty