phoenix-ralph

phoenix-ralph is a skill for Claude Code, Codex from All-The-Vibes/ATV-Phoenix. It costs 142 tokens per session (1,912 once invoked), scanned A, original, MIT.

An iteration loop that keeps an agent working through a backlog until an objective completion check is proven. Each iteration starts with fresh context and uses the filesystem to retain information.

In plain words
What is it for?
Use it for long or multi-step coding tasks that need repeated work, verification, and progress across separate agent iterations.
Why use it?
It avoids stopping just because the agent claims the task is finished. Completion requires recorded evidence that the same check failed before and passes now.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one. Also seen: mentions Claude Code.

Good fit Use it for long or multi-step coding tasks that need repeated work, verification, and progress across separate agent iterations.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/all-the-vibes/atv-phoenix/phoenix-ralph
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add All-The-Vibes/ATV-Phoenix --skill phoenix-ralph
Clone the repo
git clone --depth 1 https://github.com/All-The-Vibes/ATV-Phoenix

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for phoenix-ralph

README.md
[![agentmods](https://agentmods.dev/badge/skills/all-the-vibes/atv-phoenix/phoenix-ralph/github.svg)](https://agentmods.dev/skills/all-the-vibes/atv-phoenix/phoenix-ralph)
Your own site
<a href="https://agentmods.dev/skills/all-the-vibes/atv-phoenix/phoenix-ralph"><img src="https://agentmods.dev/badge/skills/all-the-vibes/atv-phoenix/phoenix-ralph/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for phoenix-ralph

Your own site · 80×15
<a href="https://agentmods.dev/skills/all-the-vibes/atv-phoenix/phoenix-ralph"><img src="https://agentmods.dev/badge/skills/all-the-vibes/atv-phoenix/phoenix-ralph.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 142 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,912 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00142 $0.01912
Opus 5 $0.00071 $0.00956
Sonnet 5 $0.00028 $0.00382
Haiku 4.5 $0.00014 $0.00191

Measured 11d ago against content hash ea6de560748d, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

phoenix-ralph scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/phoenix-ralph/SKILL.md · 105 lines

How it starts

The opening of the file, as written. The whole thing — 105 lines — stays where its author put it; the contents beside it link to each section on GitHub.

phoenix-ralph — the persistence loop, gated by objective proof

Ralph (Geoffrey Huntley, ghuntley.com/ralph) is a dead-simple, powerful idea: run the agent in a loop with a fixed prompt and a fresh context every iteration, with the filesystem as the only memory. One task per loop. Huntley's original is literally while :; do cat PROMPT.md | agent; done.

Phoenix adds the one thing Ralph (and Claude Code's ralph, and every "autonomous" loop) lacks: an objective, tamper-evident completion proof. The loop does not stop when the agent says it's done — it stops when the driver proves the top-level acceptance check is failure-first satisfied (phoenix-mcp accept): the trace shows the check went red → green for the same check, the chain is intact, and it is green right now.

Every other persistence loop ends in an opinion ("the reviewer approved", "the model thinks it's done"). phoenix-ralph ends in evidence.

Two ways to run it

A. Interactive (inside the Copilot CLI) — the common case. You're in a live copilot session and invoke /phoenix-ralph. There is no external script. The loop runs in-session: within your agentic turn you iterate edit → phoenix_sense → heal → phoenix_sense, one backlog item at a time, and you do not say "done" until phoenix_accept (the gate ledger, an MCP tool) returns ok=true for the done-check. The agent's own tool-use loop is the loop; the skill's rules are what keep it honest. If the job outgrows one turn, the user just says "continue" and you re-read the state files and resume.

B. Unattended / large (copilot -p) — fresh context per iteration. For overnight jobs, CI, or work too big for one context window, the external driver dist/ralph/phoenix-ralph.ps1 (+ bash twin) re-invokes the agent with a fresh context every loop (Huntley's key trick against the ~150k-token degradation). Same gate, same state files — the driver calls phoenix-mcp accept instead of the in-session tool. Use this when nobody's watching or the backlog is long.

Read the full file on GitHub · 105 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 11d ago First seen · 105 lines · 142 tokens per session scan A ea6de560748d

Subscribe to this mod's changes

phoenix-ralph is a skill published in the GitHub repository All-The-Vibes/ATV-Phoenix (5 stars, last pushed 6d ago), licensed MIT. It adds 142 tokens to every session and 1,912 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories

security-review

Perform a focused security review of pending git changes to identify high-confidence security vulnerabilities with real exploitation potential. Use this skill when the user asks for a security review, security audit, vulnerability scan, or wants to check pending changes on a branch for security issues before merging.…

waybarrios/opencode-power-pack · 64 tokens

huggingface-llm-trainer

Train or fine-tune language models with TRL or Unsloth on Hugging Face Jobs, including SFT, DPO, GRPO, reward models, and GGUF conversion. Use for cloud LLM training; use huggingface-vision-trainer for vision tasks.

waybarrios/opencode-power-pack · 65 tokens

feature-dev

Guide a feature implementation through a structured seven-phase workflow with deep codebase understanding, clarifying questions, parallel architecture design, and quality review. Use this skill when the user asks to build a new feature, add functionality, or wants a methodical approach to implementation rather than…

waybarrios/opencode-power-pack · 62 tokens

semgrep-rule-variant-creator

Creates language variants of existing Semgrep rules. Use when porting a Semgrep rule to specified target languages. Takes an existing rule and target languages as input, produces independent rule+test directories for each language.

waybarrios/opencode-power-pack · 50 tokens

learning-quality

Structured format for capturing high-quality learnings during ClosedLoop runs.

closedloop-ai/claude-plugins · 15 tokens

plan-validate

Deterministic plan.json validation via Python script, replacing most plan-validator agent calls. Performs JSON parsing, schema validation, task checkbox regex, required section checks, sync validation, and data extraction. Only semantic consistency checks (storage/query alignment) require the LLM agent. Triggers on…

closedloop-ai/claude-plugins · 89 tokens