Agent skills · by Uberblick

Agents need feedback, not more instructions.

Skills for teams that run coding agents on their GitHub issues. They read what the loop already records, report what happened, and propose the smallest change that fixes it.

npx skills@latest add uberblick-ai/skills

/plugin install ub-agents --marketplace uberblick-ai/skills

Browse the skills →

2skills

2groups

1command to install

0services to run

A loop that needs you for every item isn't a loop.

Coding agents claim issues, open pull requests and review each other's work. When a day goes well, nobody looks. When it doesn't, the cost hides in retried runs, sent-back PRs and items waiting for a person.

The evidence is already on GitHub: claims, outcomes, denied commands, retrospectives. Reading it by hand takes longer than the runs did, so it rarely happens.

These skills read it for you. They name the cause, count what it cost, and propose a fix that removes text or scope instead of adding another rule.

The skills, grouped by what they work on.

Each group is a folder in the repository and a plugin in Claude Code. Open a skill for install steps and its report.

UB Agents

plugin ub-agents

Skills for repositories that run the ub-agents loop: reading its records on GitHub, judging how the loop did, and tuning the instructions it runs on.

ub-agents-deliver-reportOne day of loop deliveries: what the loop delivered on its own, which runs were wasted, where a person stepped in, and the few changes that would have saved runs.Ask: "How did the loop do yesterday?"v1.0

Maintenance

plugin maintenance

Repository maintenance skills, independent of a delivery loop or release process.

backlog-groomingReview all issues, a milestone or a project against what already shipped. Challenge duplicates and excess scope, settle decisions in chat, and get a report of what to cut.Ask: "What can we drop from this milestone?"v1.1

What do you want to do?

Find your question, then the skill or page that answers it.

Measure the loop

How did the loop do yesterday?

Skillub-agents-deliver-report
DocsRetrospectives

Cut scope

What can we drop before the next release?

Skillbacklog-grooming
DocsSmall issues

Run the loop

How do I let agents work my issues?

DocsInstallation
DocsConfiguration

Contribute

How do I write a skill for this repository?

GuideAGENTS.md
SpecAgent Skills specification

Recent changes.

See the changelog

2026-10-10Siteskills.uberblick.ai, one page per skill.
2026-10-09SkillAdded backlog-grooming 1.1 in the new Maintenance group.
2026-10-09SkillAdded ub-agents-deliver-report, moved here from uberblick-2 and ub-agents.
2026-10-09PluginOne Claude Code plugin per skill group. The old /uberblick plugin is gone.
2026-03-07Releasev1.0.0, the first public release.

Who builds this.

Uberblick builds ub-agents, the loop engineering framework for GitHub. The skills here come from running that loop on our own repositories, ub-agents included. The deliver report screenshots are one of its days.