Skip to content
Menu

Look

Language

GitHub — @IchiSieben
Skills

LLMs & agents

AI & machine learning

Language models orchestrated as agents, with evaluation built into the flow, not the pitch.

Agents with a falsifiable prediction ledger: evaluation lives in the flow, not in the pitch.

Where it was used 03

  • enjambre

    Headless Claude Code with an explicit tool allow-list, no --dangerously-skip-permissions and a task contract (allowed files and done criteria).

  • Claude Kit

    Six subagents by lane (haiku to opus) and one review rule: the reviewer is never weaker than the author.

  • Signal Agent

    The judging step: the model reads the briefing and the criteria, emits a verdict as JSON, and eight hard rules decide whether it is allowed to persist.

Live · Usable

Agents & AI automation

enjambre

I spread Claude Code jobs across my PC and a Mac mini; each one returns as a branch I review, with a live dashboard of the machines.

Python · Bash · SSH · git worktrees · tmux · Claude Code

Case study
Live · Usable

Agents & AI automation

Claude Kit

The Claude Code setup I built this portfolio with: rules, agents, hooks and a status line that does not lie, the same on every machine.

Bash · jq · git · Claude Code

Case study
Signal Agent
In development · Prototype

Agents & AI automation

Signal Agent

A market monitor where the model has to write a falsifiable prediction before it can be scored — 3 open, 0 resolved, the first ones due January 2027.

Python · GitHub Actions · SQLite

Case study