Loop Step
Perform exactly ONE Ralph loop increment interactively (do one step → verify → log → commit → emit a marker), following ralph/PRO…
Claude CodeGeneric
---
description: Perform exactly ONE Ralph loop increment interactively (do one step → verify → log → commit → emit a marker), following ralph/PROMPT.md.
---
Execute **one** iteration of the Ralph loop by hand, exactly as
[`ralph/PROMPT.md`](../../ralph/PROMPT.md) specifies. This is the interactive
equivalent of one `ralph/loop.sh` cycle — use it to step through carefully or to
test the loop before running it unattended.
Do this now:
1. Load context in order: `CLAUDE.md` → `AGENTS.md` → `ralph/STATE.md` → the
active task folder (`BRIEF`, `PLAN`, tail of `LOG`). Confirm `git status` is
clean.
2. Respect the gates (AGENTS.md §6) — especially: if Phase 0 is open, do not
write any `bot/` code. Never cross a human boundary (live mode, real funds,
secrets, milestone tag) autonomously.
3. Do the **single next unchecked step** in the active `PLAN.md`. Just one.
4. Verify it (AGENTS.md §5) and record the evidence in `LOG.md`.
5. Check the box in `PLAN.md`; update `ralph/STATE.md` (next step, iteration +1,
gate); update `tasks/INDEX.md` if status changed.
6. Commit (Conventional Commits + scope, no ticket IDs) and push.
7. End with exactly one marker on the last line: `<<RALPH:CONTINUE>>`,
`<<RALPH:PHASE_COMPLETE>>`, `<<RALPH:GATE_FAILED>>`, or `<<RALPH:BLOCKED>>`.
Then STOP. Do not start the next step — that's the next iteration.
If `$ARGUMENTS` is given, treat it as a focus/override for which step to do, but
still do only one.
Maintain Loop Step?
Let people know it's listed here — add the badge (live metrics, light/dark aware) or a plain link to your README or docs.
[Loop Step on getagentictools](https://getagentictools.com/loops/serhiiromaniuk-loop-step?ref=badge) npx agentictools info loops/serhiiromaniuk-loop-step The second line is the CLI lookup for this page — handy in READMEs and docs.