skills that compound

The skills write themselves.

Start with 13 open-source skills, then let your team’s library grow itself. ZeroShot watches how your team actually works, identifies the patterns you keep repeating, builds new skills tuned to you — and makes sure they actually get used.

weekly skill PR · ~/api · mon 9:01am
zeroshot · opened PR #12: 3 new skills from last week’s sessions
/payments-reviewfound in 11 PRs re-flagging the same retry conventions
/migration-guardfound after 3 migrations shipped without a rollback plan
/export-perffound across 6 tickets about slow CSV exports
backtested against last week’s sessions — est. −290k tokens / wk if merged.
─── you · 4 minutes later ───
gh pr merge 12 --squash
merged · the library just got smarter. nobody wrote a wiki page.

every monday. skills are files in a PR — review them like anything else your team ships.

[1] · the loop

Find, build, and use skills — automatically.

No blank page, no prompt library in a Notion doc nobody opens. Install the CLI and the loop starts running — you start with ours, we learn yours, the library compounds.

[1]Start with our skills

Install the CLI and ship with 13 open-source skills — a full spec-to-verify workflow, day one.

[2]We identify your patterns

ZeroShot watches every session and PR review, spotting the work your team keeps repeating.

[3]We build & sharpen skills for you

Those patterns become new skills — and sharpen the ones you have. Your library compounds; tokens drop.

[2] · proprietary signal extraction

We audit every session.
50+ signals, automatically.

ZeroShot reads every Claude Code, Codex, and Cursor session — across you and your whole team, single-player or multiplayer — and runs the proprietary BuildBetter signal pipeline over it. 50+ signal types surface the issues, blockers, repeated patterns, and inefficiencies hiding in how your team actually codes.

issues & blockers

Retry loops, tool failures, and missing context — the moments your agents get stuck.

repeated patterns

The same steering and corrections, over and over, across the whole team.

inefficiencies

Token-heavy flows and re-derived context that quietly burn your budget.

conventions

The “actually, we do it this way” your reviews keep repeating.

[3] · skills that learn

Your junior inherits
your senior’s wins.

Every PR review your team writes, every steering correction, every “actually, we do it this way” becomes a skill the next person uses. The agent shows up already knowing the conventions your seniors settled on — so nobody spends tokens learning them the hard way again.

  • Conventions encoded once, applied by every agent
  • Skills update from your real review patterns, not a template
  • Draft reviews with /bb-review — suggestions cite prior conventions, cutting PR review time by 67%
  • We open a weekly PR suggesting new skills from your sessions
team conventions · /bb-review · updated 6h ago
draft my review with /bb-review
posted to PR #3308 · 3 suggestions, all citing your team’s prior conventions
─── the conventions it cited ───
Use RetryablePolicy — don’t add a new retry abstraction.
flagged by nikhil on PR #408 · 6 weeks ago
conversation is human-only. Agent traffic goes in agent_logs.
flagged by spencer on PR #3308 · 4 months ago
Migrations always include a rollback plan.
flagged by maya on PR #2089 · 8 months ago
#Each correction your team makes once is in the skill the next person uses.

[4] · highest-roi skills

We build the highest-ROI skills —
and prove it on your history.

ZeroShot ranks every skill opportunity by impact, builds it, and backtests it against your existing sessions — so you see exactly which skills speed up development the most, and save the most, before you ship them. Estimates, not vibes.

SUGGESTED OPTIMIZATIONS
add /payments-review skill~140k tok/wk

Your team re-flagged the same 3 conventions across 11 PRs.

prototype UI in HTML first~90k tok/wk

4 agents rebuilt the same dashboard 3× before settling on a layout.

resume, don’t re-explain~60k tok/wk

Cross-machine context re-derivation showed up in 6 sessions. · applied

LEFT ON THE TABLE~290k tok/wk

note ─ we open a PR every Monday with new skills from last week’s sessions. estimated savings — your mileage will vary.

+ Found for your team — skills ZeroShot wrote by watching how you work

/payments-review
from11 PRs
backtest~−140k tok/wk
used9× since monday
NEW THIS WEEK
/migration-guard
from3 rollback-less migrations
fires ondb/ changes
used4× since monday
NEW THIS WEEK
/export-perf
from6 slow-export tickets
fires onexport flows
used0× — see [5]
SKIPPED · ENFORCE IT

└ every skill is a file — review it, edit it, delete it. nothing is a black box.

[5] · enforcement & adoption

A skill no one uses is worthless.

You can have a million skill files — if no one runs them, nothing changes. ZeroShot lets you dictate when each skill fires, and tracks coverage so adoption actually happens.

  • Run skills automatically on a flow, or leave them manual — you decide
  • Require a skill before a PR goes up for review
  • Coverage tracking shows who’s actually using each skill — and who isn’t
skill enforcement · PR #412 · tue 2:14pm
git push
zeroshot guard · checking required skills for PR #412
/bb-review · required before review · ran
/migration-guard · on db/ changes · ran
/export-perf · on export flows · not run
push blocked — 2 / 3 required skills ran. export-perf was skipped.
─── 3 minutes later ───
/export-perf
skill ran · checks pass · PR #412 unblocked

why this exists

"One of my engineers spent $40,000 on tokens last month, and I genuinely don’t know whether I should stop him."

Vitaly Gordon, CEO, Faros AI · via TechCrunch

ZeroShot flagged that my Playwright skill was burning about $4k a month. It rebuilt the skill around it — now it's about $200 a month and 10× faster.

Spencer Shulem, CEO

We open-source our skills — browse the public library at skillrank.dev →

ok. give it to me. two ways in:
Download for macOSv0.10.0 · 13.4 MB · free, no account
Book a demo for your team →SOC 2 · can run locally