Open source · lives in your coding agent
Nosy keeps the record of what your team actually shipped: from git, against the decision or request it was for. You place bets. Every week, Nosy scores them.
On a real repo · metabase/metabase · last 90 days
feature requests opened
shipped with a PR that links them (plus 1 partly)
open requests with no decision yet: no assignee, no milestone, older than two weeks
headline shipped count withheld: some work is linked only in comments, so the number would undercount.
Public issues and PRs only, counted 28 Sep 2026.
Before / after
Ask your agent
It reads the code as it is today and tells you what it finds. It doesn’t know what merged on Tuesday, which request that PR was for, or that you said “two days” three weeks ago.
Ask Nosy
#412 Bulk export · shipped 24 Sep
PR #431 closes it
bet nb-260902-bulk-export
estimated S, took 9 active days (M)Nosy isn’t smarter than your agent. What it has is the record. (Illustration: product, numbers and ids are made up.)
The first five minutes
/nosy:move-inFinds your integration branch (the one PRs really merge into), your decision log or issues, and how your team links work. Once.
/nosy:shippedWhat reached that branch, when, and which request or decision each change was for. Linked only where a PR, commit or issue says so.
/nosy:bet"Bulk export, S, because 30 customers asked." Nosy prints an id; your agent puts it in the PR it writes.
/nosy:scoreNext week: which bets landed, how long they really took against the estimate, what got reverted or patched.
/nosy:teaOne page you can send. The scoreboard, every receipt, no dashboard to babysit.
House rules
A change counts as shipped only when a PR, commit or the issue itself links it. Nothing is matched by title. Incomplete counts are withheld, not shown.
Medians and estimate scores stay hidden until there are enough bets to mean something.
No comments on style, bugs or quality. Ever. Nosy talks product.
Reads code, git and issue metadata. Never customer data, secrets or keys. Writes only to your pm/ folder. Never pushes or posts on its own.
We peeked over the fence · 42 products
Claude Code, Cursor
Good at: Reads today’s code and answers well. In a blind test it matched Nosy on answer quality.
Missing: Starts from zero every session. Doesn’t remember what merged Tuesday, which request it was for, or that you said "two days".
Unblocked, DeepWiki, Dosu
Good at: Great maps of the codebase and its history, served to engineers and agents.
Missing: A map, not a verdict. No bets, no "shipped against what we decided", no score.
Spec Kit, Kiro, Taskmaster, CCPM
Good at: Turn a decision into specs, tasks and PRs. CCPM even uses git and issues as the source of truth.
Missing: They go forward from the decision. Nobody closes the loop: did it land, how long did it take, was the bet right?
Productboard Spark, Linear
Good at: Research, specs and roadmaps inside a full PM suite, with AI agents on top.
Missing: SaaS seats, and they read tickets, not what actually reached the branch.
Swarmia, Jellyfish
Good at: Strong at reading delivery from git for engineering leaders.
Missing: Priced per developer, built for managers. No product bets, no decision page.
free, MIT, in your agent
The record of what landed, against which decision, and a weekly score for every bet. Memory lives in your repo, not in a session.
Nosy was here.
Based on public docs, changelogs and pricing pages, Sep 2026. Our own column is scored by us, so read it with that in mind.
Install
Claude Code first. Codex, Cursor, Gemini CLI via npx skills add. Or no agent at all, from the CLI or a weekly GitHub Action.
Early and not published yet. Leave an email and Nosy knocks once when it is.
Nosy Cloud is coming later: one shared record and score for the whole team. The plugin stays free and open source.
$ /plugin marketplace add nosy
$ /plugin install nosy@nosy
$ /nosy:move-in
Psst. Integration branch is `develop`, not main.
Decisions: GitHub issues labeled `decision`.
Nosy was here.