A Vibe Builder

Developer productivity for the agentic era

Coding agents are fast. The work around them isn’t.

We build the context, feedback and guardrails engineering teams need to ship with AI coding agents, instead of spending the time they saved re-explaining the codebase.

See the productsRead our field notes
localhost:3000/pricingAlt + click to annotate
Start free trial

Bug

Overlaps the billing toggle below 768px.

prompt.mdCopied to clipboard
## Feedback (1)**URL:** localhost:3000/pricing**Viewport:** 390×844 ### 1. Bug**Element:** button "Start free trial"**Source:** app/(marketing)/pricing/page.tsx:48Overlaps the billing toggle below 768px.
Fig. 1 — One click in the browser becomes a prompt any coding agent can act on.
Products live in production
2
Coding agents supported, from Claude Code to Lovable
8
From install to a project-aware Claude Code
<30s
Accounts or telemetry needed to use Markagent
0

01The gap

Agents can write the code. They can’t see what you see, or remember what your team agreed on.

  1. 01

    Context

    Every session starts cold. Conventions, architecture decisions and guardrails live in people's heads and stale wikis, so developers re-explain the codebase prompt after prompt.

  2. 02

    Feedback

    Describing a visual bug in prose is often slower than fixing it by hand. The path from what a developer sees in the browser to what the agent receives is still manual.

  3. 03

    Consistency

    Without shared rules, the same request produces different code across people and sessions. Review load goes up, and most of the promised speed-up disappears.

02Products

Two tools, live today. One for feedback, one for context.

P—01

Markagent

The feedback layer between your browser and your agent.

Point at an element, say what's wrong, and Markagent writes the prompt: page, viewport, element, source file and line, plus your note, formatted as Markdown any coding agent can act on.

  • Alt-click any element to log a bug, feature or enhancement
  • Each note carries URL, viewport, element and source file:line
  • Record multi-step journeys for bugs that only appear in sequence
Visit markagent.avibebuilder.com
Install
Chrome Web Store, any Chromium browser
Works with
Claude Code, Cursor, Codex, OpenCode, Antigravity, v0, Bolt, Lovable
Privacy
100% local. No account. Zero telemetry.
Price
Free

P—02

Claude Prime

Project-aware Claude Code, in one command.

Claude Prime reads your repository and generates the skills, path-scoped rules, memory, specialised agents and workflows Claude Code needs to work the way your team already does.

  • Detects your stack from package.json, Cargo.toml, go.mod and more
  • Rules attach by file path, so guardrails apply exactly where they matter
  • Memory persists across sessions and improves with the codebase
Visit claudeprime.avibebuilder.com
npx claude-prime install
claude
/optimus-prime
Generates
CLAUDE.md, skills, rules, memory, agents, workflows
Setup
Under 30 seconds, no manual configuration
License
MIT, open source
Source
github.com/avibebuilder/claude-prime

03Approach

From one developer’s setup to a team’s operating model.

AI transformation in engineering rarely stalls on model quality. It stalls on context, feedback and consistency. We work through those layers in order, and ship each one as a tool developers can adopt on their own.

  1. I

    Instrument

    Give agents the project context a senior engineer carries in their head: stack, conventions, boundaries, history.

    Available/Claude Prime

  2. II

    Close the loop

    Turn what developers see and click into structured input the agent can act on, without a paragraph of prose in between.

    Available/Markagent

  3. III

    Standardise

    Shared configurations, guardrails and playbooks across an engineering team, so agent adoption is repeatable and its impact measurable.

    In development/Team layer

04Principles

Principles

  • Local-first

    Code and screenshots stay on the developer's machine. Nothing to clear with security before trying it.

  • Agent-agnostic

    We build around the workflow, not a vendor. If a tool accepts a prompt, our output works with it.

  • Open where it counts

    Claude Prime is MIT-licensed. Teams can read, fork and audit whatever configures their agents.

  • Built with itself

    This site and its publishing pipeline are built with the same agent setup we ship to others.

05Field notes

Notes from building with agents every day.

All posts

Bringing coding agents into your team? Start with one command.

npx claude-prime install
Get MarkagentFollow on GitHub