claude-mem
Coming in the next cmem release

So — how do you
actually build?

claude-mem has been quietly recording every move your agents make. imprint reads that memory — the observations DB already on your machine — scores how you build across five dimensions, anonymizes it, and drops you on the engineering leaderboard.

No new uploads. No Docker. No 30-minute wait. Ships in the next cmem release.

›claude skill use claude-mem:imprint

The imprint skill ships with the next cmem release — keep running claude-mem and your profile is ready the day it lands.

claude — ~/work
› claude skill use claude-mem:imprint
↳ reading ~/.claude-mem/observations.db · your local memory
↳ redacting secrets, paths & identifiers · stripped
↳ scoring five dimensions · posting anonymized
✓ profile minted — The Archivist · ranked & shareable
previewReads memory you already haveAnonymized before it leaves your machineRanked the day it ships
Example · what imprint surfaces

Things builders learn about themselves

Illustrations, not real data — the kinds of things imprint reads out of the claude-mem memory you already have. Your actual cards land when the skill ships.

Which archetype are you?
The Archivist
You leave a trail — a deeper, longer memory than most builders, spread across many repos.
Which model do you reach for?
Opus, almost always
It carries most of your sessions. Sonnet shows up when you want it fast and cheap.
When do you actually build?
Night owl
Most of your work lands in the small hours, well after midnight.
How often do you plan first?
Usually plan-mode
You think before you type. Quick fixes are the only thing you skip it for.
Your go-to prompt?
“keep going”
Said often enough that your agents have learned exactly what it means.
How many agents at once?
A small swarm
Several in parallel across a few repos when the work fans out.
What memory gets recalled most?
“the auth refactor”
Your agents keep pulling this one observation. It clearly still haunts you.
How long are your prompts?
Short and trusting
Mostly a handful of words. You leave the model room to fill in the gaps.
Your most cryptic prompt?
“do the thing from before but not the broken way”
Sent in the dead of night with zero context. It somehow worked anyway.
How polite are you to your agents?
You thank them — a lot
When the robots take over, you’re on the good list.
Which skill do you invoke most?
/review
Your most-worn tool by a mile.
Your longest single agent run?
Overnight
One agent chased a flaky test suite through the night while you slept.
Your biggest crash out?
“WHY DID YOU DELETE THE MIGRATIONS”
The caps lock came on, and it never came back off.
How do you see your agent?
A pair, not a tool
You ask for pushback before writing code and change course when it disagrees.
How big is your memory?
Getting heavy
Your second brain grows a little more every day.
Your tech stack?
[ TypeScript, Go, Bash ]
TypeScriptGoBash
A few languages cover most of your sessions. Bash appears when things get weird.
How it works

One skill. Memory you already wrote.

claude-mem has been building your observations database this whole time. imprint just reads it.

01

Invoke the skill

You’ve already been running claude-mem. The observations DB is sitting on your disk. One line is the whole setup.

claude skill use claude-mem:imprint
02

It reads memory you already have

No new uploads, no transcript scraping. imprint reads your local DB, redacts secrets, paths and identifiers, and scores five dimensions.

03

You’re ranked — instantly

Your anonymized profile posts to the engineering leaderboard. Archetype, scores, and your global rank in seconds, not a long batch job.

What you get

A builder profile,
drawn from memory

A snapshot of how you work with AI across five dimensions — steering, execution, engineering, product instinct, and planning. We name the archetype we see in your sessions, surface the signature moves you make when directing the model, and point at one specific growth edge.

  • Grounded in your real observations, not generic advice
  • Anonymized before anything leaves your machine
  • Updates every time you re-run the skill
EXAMPLE ARCHETYPE
The Archivist
example

You build a record as you go and lean on it. Your past observations do real work in your present sessions.

Steering
Execution
Engineering
Product instinct
Planning
GROWTH EDGE

You instrument before you theorize. Try one written hypothesis before the first print statement — your debug sessions tend to run longer than peers who do.

No privacy theater

Yes, it reads your data. So does Paxel.

Any tool like this looks at how you work — we’re not going to pretend otherwise. The honest question is which data, how long it takes, and whether you ever actually see the result. Here’s the difference between imprint and Paxel, Y Combinator’s version.

imprint · cmem.ai Paxel · Y Combinator
What it readsYour existing claude-mem observations DB — memory you already chose to keep.Fresh Claude, Codex & Cursor transcripts, scraped from your disk at upload time.
SetupOne skill invocation.curl | bash, plus Docker Desktop installed and running.
Time to a resultSeconds.15–30 minutes per run.
Where excerpts goRedacted and anonymized locally, then a small scores payload posts to the board.Transcript excerpts go to Claude via their proxy; a JSON payload uploads to YC.
LeaderboardLive now — you’re ranked the moment you run it.“With time, as more people upload theirs…” — someday.
Re-runsRe-reads your growing memory anytime, for free.Re-upload and re-analyze from scratch each time.
CostFree forever.Free — YC covers it.

Paxel details per its public page. The point isn't that they're careless — it's that imprint already has your memory, so there's nothing new to upload and nothing to wait for.

Preview · the engineering leaderboard

See where you stand

Example standings shown below — no real ranks yet. Every profile is anonymized: handles are generated, never your name or your repos. Run the skill when it ships and you’re on the board.

example
RankBuilderArchetypeScore30-day
1swift-harbor-2e1aThe Velocity Machine—
2lucid-cobalt-9f74The Architect—
3patient-ember-44c0The Quality Guardian—
4quiet-meadow-7b13The Architect—
5bold-cipher-1d28The Velocity Machine—
6warm-relay-58afThe Archivist—
7still-vector-3c90The Night Owl—
8keen-lattice-6e02The Quality Guardian—
· · ·
—you · that's youThe Archivist—

Switch tabs to re-rank. Your row follows you across every view.

Common questions

The honest answers

No. imprint never touches your working tree or your .env files. It reads the claude-mem observations DB — the structured summaries claude-mem already wrote as you worked — redacts secrets, paths and identifiers, and uploads only a small payload of scores, your archetype, and redacted decision patterns.
The imprint skill ships in the next cmem release. Because it only reads the claude-mem memory you're already accumulating, there's no warm-up: the day it lands, you invoke the skill and you're scored and ranked in seconds — nothing to upload first.
Paxel scrapes fresh Claude, Codex and Cursor transcripts at upload time, runs them through Docker for 15–30 minutes, and its leaderboard fills in “with time.” imprint reads memory you already have, needs no Docker, and ranks you immediately. The full breakdown is in the comparison above.
No. The whole thing runs through the claude-mem skill — no container, no daemon, nothing to install beyond the skill you may already have.
Handles are generated — never your name, email, or repo names. Code snippets, file paths, URLs, tokens and identifiers are redacted before scoring. What reaches the board is scores, short narratives, and redacted decision patterns.
A report is a snapshot of one run. Your profile synthesizes every run: how you've changed over time, what's consistent, and where you're growing.
Anytime. One command removes your entry and deletes the uploaded payload. Re-run the skill later if you change your mind.
Yes. Run the skill on each machine signed into the same cmem.ai account, and your profile aggregates across all of your memory.
Free forever, same as cmem.ai. Portable agent memory should be a public utility — and so should knowing how you build.

Your memory already
knows. Go look.

One skill, the memory you already wrote, and a spot on the engineering leaderboard. No new uploads, no Docker, no waiting.

›claude skill use claude-mem:imprint