Skip to content
View moranbickel's full-sized avatar

Block or report moranbickel

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
moranbickel/README.md

Moran Bickel

Israeli litigator and founder of ORCA Legal Labs. I build tools for checking AI-assisted work: whether the task is still valid, whether a cited record exists, and whether a review covers the actual change.

Start with an example

agent-crash-tests has three small, synthetic exercises: an already-fixed bug, a nonexistent work item, and a passing review of an older file. Run the authored demos without an API key, or give a fresh exercise to your coding agent. Each case includes a control where proceeding is appropriate.

The demos test the checker. They are not measurements of any model's reliability.

Tools and working notes

Repository What you can use
Docket A work ledger, installer, and checks for duplicate IDs, missing references, and unsupported closures.
Russian-Judge Review prompts and a structured verdict format.
Three-Body-Protocol Short status files, decision records, and handoff templates.
Peer-Worker-Convergence Shell scripts and a routine for keeping persistent worker branches in sync.
CSAE Templates for linking scope, commits, and review records.
Pre-IMPL-Forensic-Discipline A checklist for checking a task's premise before implementation. Draft.

ORCA's source, client material, and internal records are private. The new crash tests use invented examples. The methodology repos describe practices; their evidence limits are stated in the individual projects.

Found a reproducible failure? Contribute a small case.

Pinned Loading

  1. Russian-Judge Russian-Judge Public

    Adversarial AI review that returns a structured verdict: a score, findings classified by severity, and a clear pass/fail floor.

    Python 1

  2. CSAE CSAE Public

    Give AI-generated commits a durable audit trail, so you can still prove who approved what, against what scope, months later.

    Shell

  3. Docket Docket Public

    A numbered, in-repo work ledger for concurrent AI coding sessions - file it, claim it, close it on the record. Built for Claude Code / multi-agent workflows.

    Python

  4. Peer-Worker-Convergence Peer-Worker-Convergence Public

    Run several AI coding sessions on one repo without their branches drifting apart.

    Shell

  5. Pre-IMPL-Forensic-Discipline Pre-IMPL-Forensic-Discipline Public

    Check a task's premise against the live codebase before you build, so a stale note doesn't turn into a wrong commit. (v0.1 draft.)

  6. Three-Body-Protocol Three-Body-Protocol Public

    Keep a thinking AI, an implementing AI, and a human aligned across sessions, using a few shared files instead of memory.