LOOP — repeat a prompt or slash command at a fixed interval
/loop INTERVAL [–count N] PROMPT /loop INTERVAL [–count N] /COMMAND [ARGS…]
/loop repeats PROMPT or /COMMAND every INTERVAL for up to N iterations, blocking the REPL until finished or cancelled. It is designed for workflows like “check the build every 5 minutes” or “run /git status every 30 seconds while I refactor.”
A single Ctrl+C cancels the current iteration and any pending sleep, then returns to the Harvey prompt.
INTERVAL Required. Parsed by parseDurationString: a plain integer is treated as seconds (e.g. 300 → 5 minutes); Go duration strings such as 30s, 5m, and 1h30m are also accepted. Must be positive.
–count N Optional. Number of iterations, integer in [1, 100]. Default: 10.
PROMPT Free text sent as a chat turn each iteration. The same RAG augmentation, tool-loop execution, and recording that apply to normal chat apply here.
/COMMAND A slash command dispatched each iteration, exactly as if typed at the prompt. The command’s own safe-mode checks, audit logging, and recording are preserved. /exit, /quit, and /bye are recognised and stop the loop rather than exiting Harvey.
Chat iterations use the same model call as normal chat — same RAG augmentation, same tool-loop-or-plain-chat branch, same stats recording and Fountain recording — so looping a prompt behaves identically to typing it by hand repeatedly. Two things are deliberately excluded:
Interactive write-offers — the fenced-code-block “write to file?” prompts and autoExecuteReply are skipped, because an unattended loop must never block waiting for stdin input.
Skill auto-trigger — the trigger-word dispatch that redirects prompts to registered skills is skipped, because a looped prompt should reach the model directly and consistently on every iteration.
A transient error in one iteration (e.g. a model timeout) is printed inline but does not stop the loop — only Ctrl+C or the count limit does.
Check the build every five minutes, up to the default 10 times: /loop 5m Check the build and summarise any failures.
Run git status every 30 seconds, 3 times: /loop 30s –count 3 /git status
Ask the model to review recent log entries once per minute, 20 times: /loop 60s –count 20 Summarise any new errors in the log.
/help pipeline — chain prompts with confidence gating /help run — execute shell commands from the REPL