Skip to content

Luciazero 2.4.3 — Make coding agents prove their work

Choose a tag to compare

@github-actions github-actions released this 24 Aug 10:59
· 344 commits to main since this release

Make coding agents prove their work

Luciazero is a local-first verification and handoff layer for Claude Code,
Codex CLI, and compatible skill runtimes.

It adds a plan → change → verify → fix loop that:

  • tracks whether the latest check is real
  • keeps scope and lessons visible
  • routes risky work to focused review
  • transfers unfinished work across sessions with an evidence-backed Relay

Try it in 30 seconds

For Codex CLI:

npx luciazero codex

Start a new session, then run $ready.

For Claude Code:

/plugin marketplace add ohm41321/luciazero
/plugin install luciazero@luciazero

Run the shipped demos without an API key:

bash docs/assets/statusline-demo.sh
bash docs/assets/relay-demo.sh

Included in 2.4.3

  • 11 on-demand skills
  • Claude Code, Codex CLI, and compatible skill-runtime install paths
  • Evidence-backed /lucia-relay handoffs with drift detection
  • A real red/green verification statusline demo

Evidence

  • Relay protocol fixture: reference 6/6; generic Markdown handoff 1/6; stale fingerprint 5/6
  • Claude snapshot (2026-08-11): Haiku +15pp; Sonnet +31pp preliminary
  • Codex pilot (2026-08-12): 5 paired tasks, 0pp observed difference, exploratory

The Relay scores are mechanical protocol checks, not model-uplift results. The
Claude and Codex measurements are model- and task-specific snapshots; see the
benchmark methodology and limitations.

Links

If your coding agent ever says “done” without proving the right check ran,
try Luciazero and tell us which failure mode it missed.