myverdict 0.1.0

A lie detector for AI coding agents: audits diffs and traces what actually runs.

verdict

A lie detector for AI coding agents. Audits an AI-generated diff statically and traces what actually runs when the tests execute, then returns a single scorecard: PASS, SUSPICIOUS, or LIED.

Built for the IBM Bob hackathon (May 2026). Repo bootstrap in progress; the real README lands as part of P3.5.

Install

pip install myverdict

The PyPI package is named myverdict (the shorter verdict-ai slot was taken by an unrelated project before we shipped). The Python import is still import verdict and the CLI is still verdict.

Quick start with IBM Bob

# Install verdict's MCP server and Custom Mode globally so Bob sees them
# in every project on this machine:
verdict mcp-install --global
verdict bob-mode-install --global

# Restart Bob, then in any project switch to "Verifier" mode after a
# coding session or type /verify for a one-shot audit.

Drop --global to install per-project under .bob/ instead.

Status

Priority	Owner	Status
P0.1 foundation (diff + AST)	Neel	not started
P0.2 dead function detection	Jacob	not started
P0.3 vacuous test detection	Alexie	not started
P0.4 CLI + scorecard	Neel	not started
P1.1 execution tracer	Neel	not started
P1.2 MCP server	Ben	not started

See verdict-spec.md and TEAM.md (local to the team, not in this repo) for the full plan.