Selected work / 2026

Built to
be used.

Open-source tools and experiments shaped by real friction. Public enough to test. Useful enough to keep.

02
Public builds
01
Active
01
Experiment

Not everything I build belongs here. Only the work with something real to show.

View / Catalogue
Work 02 Experiment 2026

multi-agent

Independent second opinions, natively.

You already have an agent defined for one coding CLI, a plain .claude/agents/*.md file. multi-agent runs that same agent on other model CLIs: grok, codex, claude. Each target executes it natively through its own agent runtime, no API keys and no re-implementation. When independent models converge on a finding you get confidence. When they diverge you get coverage.

The question

What changes when reviewers fail differently?

The experiment

Preserve the agent role. Change the model. Compare the findings.

PythonClaude Code pluginMIT Experiment

View experiment → View on GitHub ↗

“I stopped looking for the best model. I look for models that fail differently.”

  1. 01 Moves context
  2. 02 Challenges the result
  3. 03 Reduces verification cost

Separate tools exploring the same direction: native agents working together without flattening what makes them different.

The catalogue

No.ProjectPurposeStateStackSource
01 context-bridge Session continuity Active JavaScript GitHub ↗
02 multi-agent Independent review Experiment Python GitHub ↗
New work appears here when it leaves the lab. Follow the build log →

Before it earns a place here.

  1. 01

    It solves friction I have felt.

  2. 02

    It works outside the demo.

  3. 03

    Its trade-offs are visible.

  4. 04

    Someone else can use it.

Built in private. Tested in public. Kept only if useful.

What should
exist next?

Difficult systems, useful tools and ideas that deserve a working version.

Let's connect → Read the writing →