// three labs tell you to iterate — none give you the tool
Engineer your prompts.
Don't just save them.
Version with diff and rollback, iterate with Loop, and reach every prompt from your terminal, the web, or straight from your agents — Chrome soon. The prompt library built like an engineering tool, not a notebook.
CLI · Web · MCP · diff + rollback · Chrome extension soon
// loop engineering
Stop guessing which prompt is better. Loop it.
“Improve” buttons are one-shot and subjective. Loop is an iteration workbench: it generates pattern-based variants, tests them against your own agent, and saves the winner as the next version, so every iteration is recorded, diffable, and reversible.
Generate 3 variants
One click in the editor, or `ple improve` in your terminal. Variants aren't random: Concise cuts tokens, Structured adds machine-friendly sections, Step-by-step adds reasoning scaffolding.
Test with YOUR agent
Pass --exec and every variant runs through your own agent: claude, llm, or your own script. The prompt arrives as PLE_PROMPT, on stdin, and as a {} placeholder. Outputs land side by side.
Save the winner, it becomes a version
The winning variant becomes the next version with a `Loop(strategy)` changelog. Diff any two rounds, roll back a bad one. Your version history turns into an iteration log.
- 1.Open a prompt and hit Loop next to Improve.
- 2.Optionally tell it what to optimize for, like “fewer tokens” or “stricter output format”.
- 3.Compare 3 variants side by side (with token deltas), apply one.
- 4.Hit Save and the changelog is pre-filled with the strategy and rationale.
Works with any agent CLI. The variant is passed as PLE_PROMPT, on stdin, and as a {} placeholder.
// one source of truth
The same prompt, three surfaces.
Write once. Reach it from your terminal, your browser, or the web, always the same version.
ple show, copy, get --var. Raw, agent-readable output
Inject into ChatGPT, Claude, Gemini. Your library one click away
Edit, version, organize, share. Your prompt repository in the browser
// capabilities
Everything you need to engineer better prompts.
// the old way
Your best prompts shouldn't live in scattered text files.
Scattered across tools, lost versions, overwriting work that mattered.
// why ple exists
Even Anthropic's own Skills don't sync across surfaces.
“Custom Skills do not sync across surfaces… Skills uploaded to claude.ai must be separately uploaded to the API… Claude Code Skills are filesystem-based and separate from both claude.ai and API.”
Anthropic Agent Skills documentationAnthropic's ecosystem is excellent. PLE is the missing sync layer:
ple show// agent-native
Your agents read it. You version it.
Context engineering starts with prompts you can trust — and MCP itself has no history, no diff, no rollback. PLE is the record layer underneath: agents pull the current version over MCP, you keep the full engineering history, from CLAUDE.md checklists to review prompts.
Read-only by design: an agent can browse and fill your prompts, but it can never burn your AI tokens or mutate your library. Same login as the CLI, zero extra setup.
> Use my code-review prompt on this diff
⏺promptlike · get_prompt("code-review")
⎿ v3.1.0 · 2 variables · 412 tokens
⏺promptlike · fill_variables(language: "ts", focus: "security")
⎿ ready-to-use text, current version, always
Live token count and per-call cost in the editor, priced across Claude, GPT, and Gemini, plus a token delta on every Loop variant.
// build log
No testimonials yet. Here are the receipts.
Built in the open, with no invented reviews and no fake logos. The proof a new tool can honestly show is that its own code ships green and tested.
242 tests · 3 surfaces · open CLI · versioned changelog
// pricing
Start free. Pay only for the engineering.
Saving and organizing is free forever. Paid plans unlock the parts a text file can't: unlimited history, AI iteration, and team seats.
Start organizing your prompts without friction.
Unlimited prompts, unlimited history, no friction.
One shared prompt library for your whole team.
// notes
What the labs actually say about prompting.
Notes from primary sources — Anthropic, OpenAI and Google documentation — on versioning, examples and iteration. Cited, short, and no more than a few times a month. No account needed.