skills.rest
ether-moon
ether-moonCOMMUNITY
Explore all skills by this author →
GitHub

ralph

Plan and execute iterative implementation loops using fresh-context Task subagents against a declarative spec.

Updated Nov 9, 2025
Search Tags:#git#planning#execution#spec#subagent#iterative#ralph
One-click install
npx skills add https://github.com/ether-moon/skill-set --skill ralph-ether-moon
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ralph
Source: https://github.com/ether-moon/skill-set/tree/main/plugins/skill-set/skills/ralph
Command: npx skills add https://github.com/ether-moon/skill-set --skill ralph-ether-moon

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Ralph eliminates context rot and manual coordination when planning and executing implementation work by running iterative fresh-context Task subagents against a declarative spec, so each iteration starts with a clean, well-scoped prompt and progress is reliably measured.

Core Features & Use Cases

  • Fresh-context loop: Spawn a new Task subagent each iteration so no state accumulates in the agent prompt.
  • Two-mode workflow: PLANNING mode generates or refines a declarative spec; BUILDING mode performs gap analysis and closes one gap per iteration.
  • Robust progress tracking: Detect progress via new commits or spec file hashes, handle stuck iterations, and require a concrete DONE condition before looping.
  • Practical use case: Generate a spec for a new feature, confirm the DONE condition, then run sequential build iterations that implement and commit tangible changes until acceptance criteria are met.

Quick Start

Start a Ralph planning loop to produce a declarative spec for the target feature, confirm the DONE condition, and then run the build loop to iteratively close gaps.

Frequently Asked Questions about ralph

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I prevent context rot when running iterative implementation loops?▼

Fresh-context Task subagents eliminate context rot during iterative implementation loops by spawning a new subagent each iteration, ensuring no state accumulates in the agent prompt and progress is reliably measured against a declarative spec.

How do I plan and execute software features against a declarative spec?▼

Planning and executing software features against a declarative spec uses a two-mode workflow: PLANNING mode generates or refines the spec, then BUILDING mode performs gap analysis and closes one single gap per sequential iteration.

What is the best way to track progress for iterative code generation?▼

Tracking progress for iterative code generation is best handled by detecting new git commits or spec file hashes, which allows handling stuck iterations and requiring a concrete DONE condition before looping.

Does this iterative execution approach require manual coordination between subagents?▼

No manual coordination is required; this iterative execution approach automatically spawns fresh-context Task subagents each iteration, applying prompt template substitution and negotiating a concrete DONE condition before looping.

Why does my iterative software implementation loop get stuck on incomplete tasks?▼

Iterative software implementation loops get stuck without a concrete DONE condition or single-gap-per-iteration execution; resolving this requires git-based progress detection and configurable subagent settings.

Related Skills

Explore More Software Engineering Skills→
alexanderopalexanderop

using-git-worktrees

Create isolated git worktrees for feature branches with safety checks.

Community
Advanced
oozoofrogoozoofrog
1

ralph-loop

Run bounded Codex loops with completion criteria and maximum iterations.

Community
Advanced
viteinfiniteviteinfinite
3

openspec-loop

Orchestrate OpenSpec spec implementations with a fresh subagent per spec.

Community
Intermediate
wibus-weewibus-wee

pi-coding-agent

Guide development of reproducible LLM coding agents using a deterministic REPL loop.

Community
Intermediate