SkillsAdd

smithery.ai

auto-ralph

>- Decides whether an imperative coding task (fix/add/implement/refactor — en/ro/ru) should run as a Ralph Loop via /ralph-loop. Invoke when the prompt has an imperative verb PLUS multi-step or verifiable scope (tests, concrete error, named files), or on explicit triggers "ralph this", "auto ralph", "loop it", "until done". Never on questions or "just answer"/"don't loop".

First seen Mar 25, 2026

Installation

$ npx skills add https://smithery.ai
View on skills.sh ↗ Website ↗

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

ralph-plan Similar name
mastra-ai/mastra

Interactive planning assistant that helps create focused, well-structured ralph-loop commands t…

2.3K installs
ralphinho-rfc-pipeline Similar name
affaan-m/ecc

RFC-driven multi-agent DAG execution pattern with quality gates, merge queues, and work unit or…

8K installs
ralph-loop Similar name
andrelandgraf/fullstackrecipes

Complete setup for automated agent-driven development. Define features as user stories with tes…

3.6K installs
ralph-loop Similar name
giuseppe-trisciuoglio/developer-kit

Ralph Wiggum-inspired automation loop for specification-driven development. Orchestrates task i…

1.8K installs
ralph-wiggum Similar name
fstandhartinger/ralph-wiggum

Autonomous AI coding with spec-driven development. Implements Geoffrey Huntley's iterative bash…

1.3K installs
ralph-tui-prd Similar name
subsy/ralph-tui

Generate a Product Requirements Document (PRD) for ralph-tui task orchestration. Creates PRDs w…

1.2K installs

Also in this package

Other skills from smithery.ai · top by installs.

npx skills add https://smithery.ai
smithery-ai-cli 7.8K frontend-design 6.2K skill-creator 2.5K ck:ui-ux-pro-max 1.3K pdf 1.2K xlsx 966 pptx 894 docx 884 webapp-testing 1 web-design-reviewer 657

Browse all from smithery.ai

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Skill metadata

Parsed from SKILL.md frontmatter.

Declared agents claude-code

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 3,655 B
  • docs SUMMARY.md 649 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 1 installs

SKILL.md

auto-ralph

Gate skill: scores imperative coding tasks and routes score ≥ 3 to a Ralph Loop. Output: MEREU română. Input: en/ro/ru/mixed, fără întrebări despre limbă.

Contract de activare

Primul output, mereu:

(AUTONALYK) ═══════════════════════════════════
  Task detectat: [tip]
  Scor: [X]/4 → [Ralph mode / Normal mode]
═══════════════════════════════════════════════

Faza 0 — Loop activ (OBLIGATORIU)

Dacă există .claude/ralph-loop.local.md: avertizează (loop deja activ, opțiuni: așteaptă / /cancel-ralph / adaugă manual la loop-ul curent) și STOP — al doilea loop corupe starea.

Faza 1 — Scoring determinist

bash "${CLAUDE_PLUGIN_ROOT}/skills/auto-ralph/scripts/score-task.sh" "<promptul userului>"

Output: score=N matched=[...]. Decizie: score ≥ 3 → Ralph mode; score ≤ 2 → afișează bannerul și răspunde normal (fără celelalte faze). Override-uri explicite (force-on/force-off) bat scorul — vezi references/detection-rules.md.

Faza 1.5 — Explore (opțional)

Refactor mare sau cod necunoscut la scor 3–4 → rulează întâi un subagent Explore (Task tool, subagent_type: "Explore"). Template-uri de prompt: references/explore-patterns.md.

Faza 2 — Context

bash "${CLAUDE_PLUGIN_ROOT}/skills/auto-ralph/scripts/detect-context.sh" [dir]

JSON cu: git, tests (posibil NOTESTSDETECTED), errors, structure, docker, settings (din ~/.claude/auto-ralph.local.md; parametrii expliciți din comandă au prioritate).

Faza 3 — Generare prompt

Alege template din references/prompt-patterns.md (bug-fix / feature / test / refactor / general; varianta NO_TESTS când nu există teste). Include mereu promise <promise>GATA</promise> și reminder /cancel-ralph.

Faza 4 — Confirmare (SINGURA întrebare)

Arată preview-ul prompt-ului și întreabă: Execut? (max 25 iterații) [Da - Recomandat] [Modifică prompt] [Nu] (sare peste dacă auto_execute: true în settings).

Faza 5 — Execuție (prompt-file, OBLIGATORIU)

  1. Scrie prompt-ul cu Write tool în /tmp/ralph-prompt.txt.
  2. Invocă (comandă single-line):

``bash /ralph-loop --prompt-file /tmp/ralph-prompt.txt --max-iterations 25 --completion-promise "GATA" ``

DE CE: Claude Code Bash tool respinge comenzi cu newline-uri. Prompt-ul multi-line inline cauzează Bash command permission check failed.

NU FOLOSI NICIODATĂ forma inline (va eșua):

# BROKEN - newlines in $ARGUMENTS cause Bash rejection
/ralph-loop "multi\nline\nprompt" --max-iterations 25 --completion-promise "GATA"

Safety

  • Max iterations mereu setat (default 25).
  • Nu executa fără confirmare; nu cere clarificări inutile.
  • Promise onestă: instruiește explicit să NU emită GATA neverificat.

Referințe

  • references/detection-rules.md — criterii scoring + override-uri + settings
  • references/explore-patterns.md — template-uri Explore
  • references/prompt-patterns.md — template-uri prompt + exemplu complet