Search Documentation

Find and navigate to documentation pages and their content

Expertise

Codistry works out who knows what in your repo from git history alone: the areas you own, the ones you share, and the ones that belong to teammates. Then it collaborates accordingly, deferring to you where you lead and explaining more where you're new.

On by default, zero setup

The profile is mined automatically when the Context Engine builds its index for a repo. You never fill anything in: your configured git identity is the signal. Sharing the profile with the agent is controlled by userExpertiseEnabled.

How the profile is built

The Context Engine already groups your repo into named areas: clusters of files that change together, each with a plain-English name. The expertise miner rides on those areas with one streaming pass over git log, counting who committed where. No extra model calls, no per-commit analysis, and nothing to configure.

Ownership

Each area is classified as yours (you lead it), shared (you and one other strong contributor both lead), or a teammate's.

Depth

Within your areas, commit volume separates expert from familiar. An area you author almost entirely counts as expert even if it's small.

It never guesses who you are

A profile is only attributed to you when your identity is confidently resolved: either you chose it explicitly, or your configured git identity (user.email / user.name) matches at least one commit in the repo's history. On a repo you've never contributed to, the identity stays unresolved and no expertise block is shared with the agent, rather than assuming the most prolific author is you.

Wrong person?

The Settings → Expertise tab shows who Codistry resolved you as. If it's wrong (say, a work and personal email), correct it there and the profile re-mines against the identity you pick.

What the agent does with it

When enabled, the agent's system prompt carries a concise User Expertise block built from your profile:

  • On your expert areas it defers to your judgement, skips basic explanations, and stays concise.
  • On familiar areas it assumes working context but explains a little more.
  • Everywhere else it does not assume familiarity: it explains its reasoning, surfaces tradeoffs, and asks before making large changes.
  • Your most recent work is listed so it knows what you're in the middle of.
Teammates' areas stay theirs

Areas owned by teammates appear on the Expertise map so you can see the whole codebase, but they are never presented to the agent as yours.

Suggestions on the welcome screen

Alongside the profile, Codistry can suggest what to pick up next when you open a fresh chat: work drawn from your most recent commits and the areas you own. Suggestions refresh when your HEAD commit changes (one small model call), and are controlled independently with workSuggestionsEnabled.

Expertise vs. Personalization

They're different features that work together. Expertise is mined: who knows what, computed from commit history with nothing typed in. Personalization is what you tell Codistry and what it learns from working with you: custom rules, your editable worldview, and project memory.

Privacy

Stays on your machine

The profile is computed locally from git history that's already on your disk. It is stored locally, shared only with the model you chose (as part of the system prompt, when enabled), and never sent to Codistry servers.

Settings

SettingDefaultWhat it does
userExpertiseEnabledtrueShare the profile with the agent as a system-prompt block. The profile itself is always mined with the index.
workSuggestionsEnabledtruePersonalized welcome-screen suggestions drawn from your recent commits.
Related