prompt-pack

An Agent Skill — for Claude and OpenAI Codex — for breaking big, risky work into a sequence of small, self-contained prompts — each does one shippable unit, verifies itself, and leaves the app working before the next. Run them in one chat or spread across many; because each prompt is self-contained, any unit moves cleanly to a fresh chat whenever you want (or need) one. It also writes paste-ready handoff briefings to resume a chat or relay work to another tool (e.g. Codex).

It encodes a battle-tested format (refined across many real packs) so every pack is consistent, safe, and resumable.

What it does

Three modes:

1. Author a pack — turn a large request into an ordered set of self-contained prompts saved to docs/<TOPIC>_PROMPT_PACK.md. Each prompt carries its own context, file list, acceptance criteria, "what must not change" guardrails, verification matrix, and commit message. Each is independently shippable.
2. Execute a pack prompt — run exactly one prompt in a fresh chat with a disciplined loop: read-first, verify references against current code (and flag the pack for re-planning if the code has drifted far enough that later prompts are now wrong), do only the scoped change — and if a gate needs a human or real-world signal you don't have, stop and emit it, never fake a "passed" gate — verify (regression + new behavior), report, and don't commit until told.
3. Handoff — write a single paste-ready briefing (a degenerate one-prompt pack) so a new chat — or another agent — resumes with full context, including what's shipped, the landmines, and the exact next step.

The pack file lives in the repo, so work never leaves the app half-broken, spans multiple chats and tools, and survives a context limit if you hit one. The skill reads the project's CLAUDE.md / AGENTS.md / memory and bakes the right build/test commands and conventions into every pack.

Lightweight by design. Authoring a pack is one focused planning pass, and each prompt runs in its own short, fresh chat — so token use stays modest and predictable. The cost discipline is the point: no single chat has to hold the whole job.

Quickstart — try this first

You type: "This <refactor / feature / migration> is too big for one chat — make me a prompt pack." (Or, if you used ideate: "Make a prompt pack from docs/CONCEPT_BRIEF.md.")

You get back: docs/<TOPIC>_PROMPT_PACK.md — an ordered set of self-contained prompts. Run each unit — one chat or many — verify, commit, move to the next. Any step can move to a fresh chat if one fills up.

Install

This is an open Agent Skill — the same prompt-pack/ folder works in Claude and OpenAI Codex. Pick your tool:

You use…	Install it by…
Claude Code — terminal, the Code tab of the Claude desktop app, claude.ai/code, or a VS Code / JetBrains IDE	dropping prompt-pack/ into ~/.claude/skills/ (all projects) or .claude/skills/ (one project)
OpenAI Codex — CLI, app, or IDE	dropping prompt-pack/ into ~/.agents/skills/ (all repos) or .agents/skills/ (one repo), then restarting Codex
Claude chat — the Chat tab of the desktop app, or claude.ai	uploading prompt-pack.skill (the zip in this repo) under Customize → Skills
Any other agent	pointing it at prompt-pack/SKILL.md — it's just instructions

The .skill file is just the prompt-pack/ folder zipped, so one unzip drops it into either skills home:

# Claude Code — detected in-session (verify with /skills):
unzip prompt-pack.skill -d ~/.claude/skills/

# OpenAI Codex — the current skills path is ~/.agents/skills/; restart Codex after:
mkdir -p ~/.agents/skills && unzip prompt-pack.skill -d ~/.agents/skills/

Prefer a clone? git clone https://github.com/nelsonwerd/prompt-pack-skill.git, then cp -r prompt-pack-skill/prompt-pack into whichever skills folder above.

_{Menu names and exact paths shift between versions — the Claude Skills and Codex Skills docs are the source of truth. Auto-activation by description is native to Claude; in Codex you invoke the skill explicitly. The methodology is identical across both.}

Runtime support

	Claude chat	Claude Code	OpenAI Codex	Other agents
prompt-pack	Limited — best for high-level planning/handoffs; weak without repo access	Best	Best — reads AGENTS.md, full repo access	Works — with repo/file access

Use it

• Manually: type /prompt-pack and describe the big job (or paste a pack prompt to execute it).
• Automatically: Claude invokes it when you say things like "make a prompt pack", "break this into phases", "I'm running out of context — write me a handoff", or "execute P3 from the pack."

Examples:

• "This refactor is too big for one chat — make me a prompt pack."
• "Break the onboarding rebuild into phases I can ship one at a time."
• "I'm almost out of tokens. Write a handoff I can paste into a new chat."
• "Execute P2 from the currency-fix pack."

What's in this repo

• prompt-pack/ — the skill itself (SKILL.md + reference playbooks). This is what you install.
  • references/pack-template.md — the canonical fill-in scaffold for a pack.
  • references/authoring-guide.md — how to turn a big job into a pack.
  • references/execution-guide.md — how to run one pack prompt safely.
  • references/handoff-guide.md — how to write a paste-ready handoff.
• prompt-pack.skill — the same folder, zipped, for a one-step download.