From ralph-harness
Use to report the status of the Ralph loop in the CURRENT project — reads the runtime state files and git, works whether or not a loop is running. Triggers: "ralph status", "how's the loop", "is the loop running", "what did the loop do".
How this skill is triggered — by the user, by Claude, or both
Slash command
/ralph-harness:ralph-statusThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Produce a one-shot digest of the loop's state for the current project. This
Produce a one-shot digest of the loop's state for the current project. This
replaces the old ralph-status.sh shell digest: read the state files and git
directly and report — no script is shipped to the project. Work whether or not a
loop is currently running.
Resolve the state dir from ralph.conf (RALPH_STATE_DIR, default .ralph) in
the repo root, then read:
RALPH_CONTAINER (default ralph-loop) and
RALPH_RUNTIME (default podman) from ralph.conf. If that runtime is on
PATH, run <runtime> ps --filter name=<container> --format '{{.Names}} {{.Status}}'.
If nothing matches (or the runtime is absent), report "not running".<state_dir>/current.json (the heartbeat: turn, task,
model, state, started). If absent, "no heartbeat yet".<state_dir>/status.jsonl; each line
is a JSON record with turn, model, exit_code, committed, sha, task. Summarize
one per line, marking committed turns.STATUS.md. Treat it as a stop reason only if it has
non-whitespace content (mirror the runner's rule — a blank/whitespace-only
file is NOT a stop). Show the reason if stopped.RALPH_REVIEW_GATE=1 in ralph.conf) — report it is
active, and surface any open review feedback: if review-findings.md exists and
is non-empty, the last review found issues the next turn must resolve, so show a
short summary (the PR number from its header + the finding count). If it is
absent/empty, the last review was clean. If the gate is off, omit this section.git log --oneline -6.Print a compact digest, in this order: running state → current turn → recent turns (newest last) → STATUS.md stop state → review-gate state (if on) → recent commits. Keep it scannable (a few lines each). Don't editorialize; report the facts. If the state dir doesn't exist yet, say the loop has not run in this repo.
STATUS.md is NOT a stop signal — never report it
as one.Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub swinney/keep-on-ralphing --plugin ralph-harness