Commands

romp is a single binary. Every command runs against the repo in your current directory (resolved from its origin remote) unless noted.

romp skills

Romp embeds the rompify skill in every binary. The skill inspects a repository and converts a short request or existing GitHub issue into a deterministic execution contract. It does not implement or run the issue.

romp skills install                   # install for detected agents
romp skills install --agent codex     # Codex CLI only
romp skills install --agent claude    # Claude Code only
romp skills install --agent openclaw  # OpenClaw only
romp skills install --agent all       # every supported agent
romp skills install --dry-run         # preview exact files
romp skills status
romp skills uninstall --agent codex

Install writes rompify/ below the selected agent’s skill directory and records the binary version in .romp-version. Running install again replaces that directory with the skill bundled in the current binary. Without --agent, install targets detected agent configuration directories. It fails without writing when no supported agent is detected.

romp run

Run one issue now, in the foreground, without claiming it in the job table.

romp run -i 17

run is a pure foreground bypass: no job row, no dedupe, absent from status, no log file. It only cleans its own stale romp-N branch and worktree.

romp watch

Poll for trigger-labelled issues and work them at width.

romp watch

On an interactive terminal, watch shows active jobs and recent history in a full-screen dashboard. Each active job exposes its current agent, verification, review, fix, or publication phase and the harness currently acting. History and job details retain both builder and reviewer harness identities. Use Tab to switch views, the arrow keys to navigate, Enter to inspect a job’s phase timeline, and Escape to return.

Successful verification output stays out of the job log. The log records the command, duration, and result, while a failed command retains its output for diagnosis. Non-interactive runs keep line-oriented output for scripts and service managers. Line output announces each read-only reviewer immediately before its process starts.

The poll predicate is “has the trigger label and does not have the claim label”. The poll interval is fixed at 60 seconds.

romp status and romp history

romp status            # in-flight jobs in this repo
romp status --all      # in-flight jobs across every repo on this machine
romp history           # recently finished jobs in this repo
romp history --all     # recently finished jobs across every repo
romp history --review --days 30

status shows each in-flight job’s codename, issue, branch, elapsed time, and optional SESSION identifier. history lists recently finished jobs with their outcome, branch, PR URL, and optional SESSION identifier. A missing session identifier appears as -.

history --review reports the clean-pass rate, fix-round rate, total re-verification failures, and median review-pass wall duration over the selected history window. One pass contains all concurrent lens calls, so the duration tracks the slowest completed lens rather than the sum of reviewer compute. It uses reviewed jobs as the rate denominator, so disabled and docs-only skips do not dilute the rates.

romp logs

romp logs sunny_naruto       # by codename
romp logs 17                 # or by issue number
romp logs sunny_naruto -f    # follow new lines

logs reads the per-job log file directly, so it works with no watcher running.

romp cancel

romp cancel 17

Cancel talks to the live watcher over a Unix socket — no watcher, nothing to cancel. Cancelling kills the agent, records a cancelled outcome, and removes the claim label, the trigger label, the job row, and the worktree and romp-N branch. The issue stays label-free until a human re-labels it.

romp gc

romp gc                 # dry-run: print what it would remove
romp gc --apply         # actually delete
romp gc --history-days 14

gc removes stale worktrees whose job has no in-flight row, and prunes finished-job history older than history_days (default 30, from user config).

romp doctor

romp doctor

Checks five things and prints a table:

• git — installed and 2.35+
• gh — installed and authenticated
• harness — at least one of claude, codex, or opencode is healthy
• config — romp.toml loads and a verify command is present
• review — the effective reviewer CLI is healthy, or review is disabled

romp init

romp init

In an interactive terminal, inspects project configuration files and lets you select or type one or more ordered verification commands. It then asks whether to enable review, with yes as the default, before writing romp.toml. It creates the three labels on the repo and appends .romp/local.toml to .gitignore. If romp.toml already exists it is left untouched. In non-interactive use, pass one or more repeatable --verify flags; Romp does not guess commands and writes review as enabled.