ralph-claude

Generate a structured requirements document for features OR bug investigations. Use when planning a feature, troubleshooting a bug, or any multi-step effort. Triggers on: create a prd, write prd for, plan this feature, investigate this bug, troubleshoot, debug.

Native-subagent alternative to /ralph-pilot. Orchestrates a PRD execution end-to-end as a single Claude Code session that preserves its context across the whole PRD by delegating EVERY implementation story to a sub-agent via the Agent tool — defaulting to opus, downgrading to sonnet/haiku only with explicit reasoning when the work is mechanical. The orchestrator's main thread is reserved for orchestration (review, apply, validate, commit) plus safety-carve-outs (decision gates, discovery stories, final validation). Backed by a systemd-user-timer watchdog that resurrects the orchestrator via `claude --resume` if it dies. Triggers on "run this PRD natively", "ralph-pilot-native", "native ralph", "skip the ralph loop", "execute this PRD with subagents".

Audit a PRD (or any plan document) with parallel disjoint-angle audit agents before running Ralph. Catches blockers, mega-stories, broken codebase assumptions, and design flaws while they're still cheap to fix. Use when a PRD has been drafted but not yet converted to prd.json. Triggers on: audit this prd, ralph audit, audit before ralph, pressure-test the prd, audit the plan.

Enable or disable cavemanMode on a Ralph PRD. When enabled, Ralph compresses progress files via caveman-compress (reducing input tokens on every iteration), runs a one-time compression pass on AGENTS.md/CLAUDE.md during /ralph-worktree setup, and uses caveman-style terse chat output when running headless. Commits, decision files, and AGENTS.md additions during a run always stay in full prose. Use when the user says 'run this with /ralph-caveman', 'enable caveman', 'turn on compression', 'disable caveman', or when /ralph-pilot is recommending compression separately from model delegation. Triggers on: /ralph-caveman, ralph-caveman, enable caveman, disable caveman, turn on compression, turn off compression.

Enable or disable modelHintMode on a Ralph PRD. When enabled, Ralph delegates stories tagged with modelHint sonnet/haiku to Agent sub-tasks on those cheaper models (Opus reviews the patch and commits), and independent unblocked stories may run in parallel. Use when the user says 'run this with /ralph-modelhint', 'enable model hints', 'turn on delegation', 'disable model hints', or when /ralph-pilot is recommending delegation separately from caveman mode. Triggers on: /ralph-modelhint, ralph-modelhint, enable model hints, disable model hints, turn on delegation, turn off delegation.

The operator role for the Ralph autonomous agent execution system. Co-authors PRDs with the user across multiple sessions, audits and refines scope before launch, sets up execution (single or multi-PRD), monitors running loops, intervenes on failures, preserves knowledge across sessions, and maintains the strategic/tracker/roadmap document set. This is the META-skill that orchestrates all other Ralph skills (/prd, /research-prd, /ralph-audit, /ralph-handoff, /ralph, /ralph-worktree, /ralph-runner). TRIGGER when the user is planning a non-trivial engineering effort that will be executed via Ralph, when monitoring a running Ralph loop, when intervening on a stuck or failing agent, when reviewing post-Ralph output, or when transitioning work across sessions. Trigger phrases include "help me plan this", "design this effort", "run this PRD", "monitor ralph", "check on ralph", "ralph status", "something broke in ralph", "resume this project", "what's next on this project", and any request that implies operating the

Convert PRDs to prd.json format for the Ralph autonomous agent system. Use when you have an existing PRD and need to convert it to Ralph's JSON format. Triggers on: convert this prd, turn this into ralph format, create prd.json from this, ralph json, start ralph.

Create an isolated git worktree for a Ralph PRD so it can run without conflicting with other work. Use after /prd and /ralph have created the task directory and prd.json. Triggers on: start ralph, create worktree for ralph, ralph worktree, launch ralph, kick off ralph.

Launch multiple Ralph PRDs sequentially or in parallel with isolated git worktrees, proper PID tracking, and --status / --stop / --clean subcommands. Use when you have two or more PRDs to execute at once (parallel when independent, sequential when dependent) rather than running a single PRD via ralph-tui or direct ralph.sh invocation. Provides the launcher script template, the PRD registry format, known pitfalls (Git Bash PID namespace, nohup wrapper PID capture, hung SSH recovery, Windows drive letter delimiter conflicts), and the relationship to /ralph-worktree (which handles a single PRD). TRIGGER phrases include "launch multiple PRDs", "run these in parallel", "run these in sequence", "launch the next batch", "launch-fips-agents", "ralph runner", "multi-PRD execution", and any request to orchestrate two or more Ralph loops simultaneously. This is the execution layer that /ralph-pilot reaches for in Phase 2 when a single worktree isn't enough.

Create or update a handoff.md document that captures the current state of a long-running project so the next agent (or the same agent in a fresh context window) can pick up exactly where this one left off. Use when context is getting tight, when finishing a session, after merging a major PRD, or whenever the user says 'update the handoff', 'write a handoff', 'hand this off', or 'create a handoff doc'. Triggers on: handoff, update handoff, hand off, write handoff, /ralph-handoff.

Generate a structured requirements document for research efforts — stories produce decision documents with citations, not code. Human decision gates between research phases. Triggers on: research prd, create research prd, plan research, research sprint.