Your library, from the terminal.

Save a prompt locally — no account needed. Body comes from -m or from stdin. Prompts live as plain Markdown files in ~/.promptlike/prompts/ (greppable, diffable, git-committable). Overwriting an existing prompt requires --force; tags and the description you don't repeat are carried over.

$ ple save "code review" -m "Review this diff: {{diff}}" -t review
✓ Saved: code review (code-review)
  ~/.promptlike/prompts/code-review.md
  variables: {{diff}}

$ pbpaste | ple save "bug triage"     # or pipe the body in

Signed out, lists your local prompts; signed in, lists every prompt your account can access (tag filter server-side, default 30, max 100; --project limits it to one repo, CLI 0.8+). Same command, different source.

$ ple list
ID        TITLE                          UPDATED
────────  ─────────────────────────────  ──────────
12ab34cd  Senior Code Reviewer           2d ago
98ef76dc  Bug Triage Assistant           5d ago

$ ple list --tag debug --limit 50

CLI 0.8+, signed in. Create and manage repos from the terminal. A new repo is private unless you pass --public. The server enforces everything: your plan's repo limit (Free: 1), name rules, unique URL slug, and that only the owner can change visibility. Making a repo public asks for confirmation (--yes in scripts). There is no delete command on purpose: deleting a repo cannot be undone, so it stays on the web. Add --json for a stable machine-readable shape.

$ ple repo create "team-prompts" -d "Shared review prompts"
✓ Created private repo 'team-prompts' (team-prompts, 1a2b3c4d)

$ ple repo list
ID        NAME                            VISIBILITY  ROLE    PROMPTS  UPDATED
1a2b3c4d  team-prompts                    private     owner         0  now

$ ple repo visibility team-prompts public --yes
✓ 'team-prompts' is now public.

Case-insensitive search across title and description (signed out it also searches local prompt bodies and tags).

$ ple search debug
2 results:

Bug Triage Assistant · 98ef76dc · 5d ago
  Triages bug reports by severity and reproducibility.

Debug Helper · 33cc44dd · 2w ago
  Walk through systematic debugging steps.

Render a prompt in the terminal with section headers. Matches by title substring or short ID.

$ ple show "code review"
── ROLE ──
You are a senior code reviewer…
── DESCRIPTION ──
Reviews diffs for correctness and clarity.
── CONTEXT ──
…

Look up a prompt by title substring or short ID and send it straight to your system clipboard. If the same title lives in more than one repo, the candidates are listed with their repo names, and --project <repo> picks one.

$ ple copy "code review"
✓ Copied: Senior Code Reviewer (12ab34cd, 487 chars)

$ ple copy 12ab34cd            # short ID also works
$ ple copy "code review" --project backend-prompts

Print a prompt raw to stdout (no colors, no decoration) so agents and scripts can pipe it. Fills {{variable}} placeholders from repeatable --var flags; messages go to stderr to keep stdout clean. Add --project <repo> to scope to one repo when the same title exists in more than one.

$ ple get "code review" --var lang=python | claude

$ ple get "code review" --project backend-prompts   # disambiguate by repo

CLI 0.5+. Turn a prompt into an AGENTS.md section (printed to stdout, so you can append it) or an Agent Skill folder with a SKILL.md that Claude Code, Claude and other skill-aware agents load on demand. Leave out the name and pass --project to export a whole repo: one AGENTS.md, or one skill with each prompt under references/. --zip writes the archive Claude's skill upload expects. Works signed out too, from your local prompts.

$ ple export "code review" --format agents-md >> AGENTS.md

$ ple export "code review" --format skill --out ~/.claude/skills
✓ Exported 1 prompt → ~/.claude/skills/code-review

$ ple export --project backend-prompts --format skill --zip

CLI 0.6+. Bring in the prompts you already keep for your agents, without copy-pasting: .claude/commands, .claude/skills/*/SKILL.md, .cursor/commands, ~/.codex/prompts, .github/prompts/*.prompt.md, AGENTS.md and CLAUDE.md (this repo and your home folder). Pick 1,3-5 or all. Re-running never imports the same content twice. Secret files (.env*, keys) are never opened, and a file that looks like it holds a live token is skipped.

$ ple import --scan --dry-run
Found 6 new · 2 already imported
  1  claude-code  command  review          .claude/commands/review.md
  2  cursor       command  refactor        .cursor/commands/refactor.md
  …

$ ple import --scan            # pick: 1,3-5 / all / Enter to cancel

CLI 0.6+, signed in. The version history from the web, in the terminal. diff compares any two versions (3, v3 or 1.0.2; default previous ↔ current) as a unified diff, colored in a terminal and plain when piped. rollback is non-destructive: the old version is copied forward as a new one, so a rollback can itself be undone.

$ ple log "code review"
* v5  1.0.4  rollback      2h ago
  v4  1.0.3  web           3d ago

$ ple diff "code review" 2 5 | delta
$ ple rollback "code review" v2 --yes

CLI 0.6+, signed in. How your library is actually used: uses in the window, where they came from (web, CLI, agents), the most used prompts and the ones not touched in 30 days. Free shows a 7-day view.

$ ple stats --range 30
Usage — last 30 days
Uses            214
Outside the web 71%
…

CLI 0.6+. Which saved prompt fits this task? Matching is local: a cached copy of your library, or your local prompts when signed out. The text you type never leaves your machine. --hook is the Claude Code mode (see below).

$ ple suggest "review this pull request for security issues"
Code Review  ai-os · used 12× · score 7.4

Use one: ple get "Code Review"

CLI 0.6+. Preferences, kept apart from the session so logging out does not reset them. Today: suggest on|off.

$ ple config set suggest off
✓ suggest = off

Upload your local prompts into your signed-in account — one-way, explicit, never automatic on login. Sync copies (local files are kept), tags travel too, and every shortfall is reported with its reason. --dry-run lists what would be uploaded without changing anything.

$ ple sync --dry-run
2 local prompts would be uploaded:
  code-review  code review
  bug-triage   bug triage

$ ple sync
✓ Uploaded 2 prompts (+3 tag links)

Sign in with email + password — this is the upgrade, not the door. Adds sync, version history (up to 3 versions of each prompt on Free; the 4th save asks you to upgrade), and the web editor. The session is stored at ~/.config/promptlike/config.json (owner-only, 0600). After signing in, commands read your account library; bring local prompts up with ple sync.

$ ple login
Email: you@example.com
Password: ********
✓ Logged in as you@example.com
2 local prompts on this machine are not shown while signed in — bring them into your account: ple sync --dry-run

Loop Engineering, single round: the API generates 2-3 AI variants of your prompt (concise / structured / chain-of-thought), each with a rationale and a size comparison against the current version. Pick a winner and it's saved as a new version with a Loop(<strategy>) changelog. Pro feature.

$ ple improve "code review" --focus "shorter"

[1] CONCISE · 412 chars · ~103 tokens (-75 chars vs current)
  why: Cuts filler, keeps the checklist.
  …
Pick a winner [1-3], 0 to cancel: 1
✓ Saved 'Senior Code Reviewer' as v5 (concise).

Multi-round improve: connect your own agent as the judge. With --exec, every variant is run through your command (prompt delivered as the PLE_PROMPT env var and on stdin; a literal {} is substituted) and the first ~20 lines of output appear under the variant. Each round's winner becomes the base for the next. Pro feature.

$ ple loop "code review" --exec 'claude -p {}' --rounds 3

═══ Round 1/3: 'Senior Code Reviewer' ═══
[1] CONCISE · …
  ── exec output exit 0 ──
  │ Here's my review of the diff…
Pick a winner [1-3], 0 to cancel: 2
✓ Saved 'Senior Code Reviewer' as v6 (structured).
Continue to next round? (y/n) y

Remove the local session file. The session itself expires on its own clock.

$ ple logout
✓ Logged out