vanguarstew
SN74 · repository-maintainer agent v0.7.0

The maintainer agent that maintains itself.

vanguarstew reviews, scores, merges, and closes pull requests on its own repository — autonomously, in public, with a human holding the keys. And it gets measurably better every week.

Autonomous
co-maintainer
Measured
perf:xs–xl scoring
Self-improving
verified against history
SN74 · MIT
on Gittensor · open source

The maintainer loop

Not "did it write good code?" — "would it have steered the repo the way a strong maintainer did?"

Given a repository frozen at a moment in time, the agent decides what a strong maintainer would do next — in four steps.

01

Infer the philosophy

Read the history, README, and recent activity to work out where the project is going. The hardest, most important part.

02

Read the situation

Open issues, open PRs, recent commits, releases — the maintainer's surface as of that moment, nothing from the future.

03

Plan & decide

Propose the next actions and the concrete calls — merge, request-changes, triage, review, release — each with its reasoning.

04

Implement

Produce an actual patch when that's the right move — but writing code is only one of the calls a maintainer makes.

Proof, not a pitch

It found a flaw in its own judgment — and fixed the system underneath it.

Run live against a real, fast-moving public repo, vanguarstew caught itself being confidently wrong — no crash, no bug report. Then it did the thing that separates a maintainer from a bot.

It kept predicting a release was imminent when the maintainers cut none. Contributors sent patches to tune the model. But reviewing them, vanguarstew reached a different conclusion: the code was never the problem — the benchmark's own yardstick was.

It was scoring over "the next 5 commits" — ~24 minutes of work on a fast repo, ~46 days on a slow one. So instead of tuning the model, it redesigned the measurement itself: each repo judged over a time window matched to its own release cadence. Fix the system, not the symptom.

Every number is checkable against real git history — that's the point of building on it.

It maintains itself

The maintainer agent, running the maintainer loop on its own repository.

The same agent the benchmark scores now co-maintains vanguarstew live — openly, as a transparent bot, with the human owner holding admin and a kill-switch.

Dogfooding, in the open — the strongest proof the concept works.

Reviews in context Judges each PR against the repo's inferred direction — a real verdict, not a rubber stamp.
Labels by measurement, not by reading the diff Runs the real benchmark against a public AND a private repo set, then bands the PR perf:xs–perf:xl by the worse of the two — never a guess.
Updates & merges green PRs Auto-updates behind branches, waits for CI, and merges only approved, conflict-free, up-to-date work.
Catches stale approvals When a landed PR breaks an older one, it flips its own approval to request-changes and drops the tier.
Keeps the queue clean Closes off-scope and stale-conflicting PRs, and rejects AI-attributed contributions — with a polite note.
Reports its day Posts a daily activity digest to a pinned issue, so every autonomous action is auditable.

How it's scored

A benchmark built from real GitHub history.

Freeze a repo at time T, let the agent plan the next moves, then judge them against what the maintainers actually did — hardened against leakage and judge bias.

Freeze · T

Rebuild the past

Reconstruct the repo, issues, PRs and releases exactly as they were knowable at T — nothing leaks from the future.

Plan

Agent decides

The agent infers the philosophy and plans the next maintainer actions with its reasoning.

Judge

Pairwise + anchor

A dual-order pairwise judge weighs trajectory and reasoning, blended with an objective anchor of hard ground truth.

Composite

One score

Judge and anchor fold into a single tunable composite — averaged over many freeze-points and repos.

The benchmark scores the agent across many repos at once, on a per-repo time horizon and human-era history. Generalization is the whole ballgame: an agent that maintains one repo is a script; one that maintains any repo is a maintainer.

A different category

Not an agent framework. Not an issue-resolver.

Measuring maintainer judgment is genuinely new — a separate skill from wiring up agents or writing a patch that passes hidden tests.

Agent frameworks"How do I wire up agents?"
MeasuresOrchestration plumbing
Ground truth— none
AnswersHow to build an agent
e.g.LangGraph, AutoGen, CrewAI
Issue-resolution benchmarks"Can it write a passing patch?"
MeasuresPatch correctness
Ground truthHidden unit tests
AnswersCan it close one ticket
e.g.SWE-bench
vanguarstew"Would it steer the repo like a strong maintainer?"
MeasuresMaintainer judgment & foresight
Ground truthReal git history, public + held-out
AnswersWhere the project should go next
e.g.this — and it's measured in public

Live evidence

Every perf:* label, measured — never claimed.

One bar per pull request, on a fixed anchor release — every real agent/ score the maintainer bot has ever published, including the ones that got closed. No line chart, no averages to squint at.

anchorv0.7.0 release
measured onpublic + private targets
label followsthe worse of the two
updatedafter every real score

Where it's headed

From one maintainer to an orchestra of agents.

Roadmap · not live yet

The judgment core proven here becomes the manager — a maintainer that doesn't just decide, but directs a team of specialists, reviews their work, and sends back rework.

The manager

vanguarstew

Sets priorities, delegates, reviews results, approves or sends rework — the maintainer's judgment, running the room.

● proven today — the core loop
Security
Audits changes and dependencies for risk.
planned
Research
Investigates approaches and prior art before work starts.
planned
Market
Tracks what users and the ecosystem actually need next.
planned
Web / dev
Implements the features and fixes the plan calls for.
planned
Survey
Gathers signal from issues, discussions, and usage.
planned
Review
Second-opinion checks before the manager signs off.
planned

Today: the maintainer judgment core, measured and proven. Next: it runs the team.

Powered by Gittensor

Built by a network that gets paid to improve real software.

vanguarstew runs on Gittensor — a Bittensor subnet (SN74) that rewards a network of contributors for making real, merged improvements to open-source repositories. Every merged PR makes the agent — and the benchmark — better.

subnetSN74 · Bittensor
releasev0.7.0
licenseMIT
nextM7 · legible foresight metric