Mine the AI prompts your team repeats. Share them as agents.
Operon reads your team's Claude Code, Codex, and Cursor sessions, surfaces the workflows that keep coming up, and turns the best into shared agents anyone can install in one command.
$ ▋
operon · scanning 1,240 sessions across 8 engineers
✓ 0 prompts recur across people and weeks
review-pr×344 engineers~25 min/run
write-migration×213 engineers~18 min/run
triage-sentry×185 engineers~12 min/run
$ ▋
✓ Claude drafted SKILL.md
✓ published acme/review-pr@1.0.0
# on every teammate's machine
$ ▋
✓ installed · works in Claude Code, Codex, Cursor
$ ▋
Your best AI work is hiding in plain sight.
AI is making a few of your engineers dramatically faster. That speed never reaches the rest of the team.
It lives with a few people
One or two engineers have figured out what works. Everyone else is still typing prompts from scratch.
It never compounds
A good prompt gets used once and forgotten, so the team keeps relearning the same things.
acme/platform · 12 agents · 16 engineers
audit-website@0.3.2 0 runs · 0 installed
release-notes@0.4.0 0 runs · 0 installed
clone-website@0.2.1 0 runs · 0 installed
pr-review@1.2.1 0 runs · 0 installed
api-contract@0.5.0 0 runs · 0 installed
✓ 0 runs this month · $0 spend
Three commands, no new habits.
It runs on the AI tools your team already uses, so there's nothing to document and no new workflow to learn.
1
Mine your sessions
Operon reads your Claude Code, Codex, and Cursor history on your own laptop and surfaces the prompts your team repeats most. That's the mining.
2
Publish the best as an agent
Pick a winner and Claude writes it a clean SKILL.md — a versioned agent that lands in your team registry.
3
Everyone installs it
One command drops the agent into any teammate's setup. It works the same in Claude Code, Codex, and Cursor.
Your whole team levels up.
Your team's best agents all live in one place.
Onboard in hours, not weeks
New hires get every team agent on day one, instead of digging through Slack.
$ operon setup
✓ detected claude-code, codex
✓ joined acme/platform
✓ installed 7 team agents
your team's top agents
audit-website 0 uses ★
clone-website 0 uses ★
setup-claude 0 uses
release-notes 0 uses
$ ▋
See which agents get used
Every run is logged: who ran it, how long it took, and what it cost.
runs
0
active agents
0
spend
$0
top by use
audit-website 0 runs · $0
release-notes 0 runs · $0
clone-website 0 runs · $0
candidate to cut
legacy-importer 0 runs · 90d idle
Knowledge stays when people leave
When someone leaves, their agents stay, and the next hire picks up where they left off.
# authors gone, agents stayed
release-notes@0.4.0 · arush.m left 9mo ago
✓ 0 runs since
pr-review@1.2.1 · priya.k left 18mo ago
✓ 0 runs since
$ ▋
Fork agents for each team
Fork it, change it, and push it back, just like any other code.
acme/review-pr@1.2.0 base · platform team
└─ acme-product/review-pr@1.2.1 + web-app conventions
└─ acme-infra/review-pr@1.2.2 + terraform & helm checks
✓ 0 forks · synced from base
$ ▋
Built like engineering infrastructure.
Each agent is versioned, observable, and scoped to exactly what it's allowed to touch.
v0.3.2 fix: tighten lighthouse threshold · 2d · arush current
v0.3.1 feat: add a11y audit step · 5d · arne
v0.3.0 feat: nightly cron · 2w · arush
v0.2.4 fix: handle 404 selector · 3w · arne rollback
✓ operon rollback v0.2.4 · restored in 0.3s
10:42 audit-website@ 0.3.2 ✓ ok 1.4s $0.012
10:31 clone-website@ 0.2.1 ✓ ok 0.9s $0.008
10:18 audit-website@ 0.3.2 ✓ ok 1.2s $0.011
10:04 setup-claude@ 0.1.0 skip 0.0s —
09:51 audit-website@ 0.3.2 ✕ err 2.1s $0.014
09:42 release-notes@ 0.4.0 ✓ ok 0.7s $0.005
0 runs · $0 · 0% ok
tools → web.fetch, lighthouse.run
permissions → read:catalog
budget → $2.00 / day
per-call cap → $0.50 · deny if over
spent today → $0 / $2.00 · 0%
✓ under budget · within scope
# on this laptop · private
~/.operon/sessions/ 0 files
~/.operon/candidates/ 0 drafts
~/.operon/cache/ 0 MB
↓ redaction runs locally
↓ you choose what to publish
# shared with team · uploaded
✓ audit-website@0.3.2 team
✓ clone-website@0.2.1 team
✓ release-notes@0.4.0 public
Ship your team's first agent today.
Free to start, no card required. Two minutes from install to your first mined prompt.
Works with Claude Code · Codex · Cursor