Analysis updated 2026-05-18
Install as a Claude Code skill to force an agent to prove claims before marking a task done
Run a structured refutation round where independent reviewers challenge finished AI generated work
Produce a machine readable evidence card documenting verdict and evidence maturity for a task
Validate an evidence card file against the project's schema before sharing it
| voltrace-io/vibes-to-verified | 0xallam/my-recipe | 0xhassaan/nn-from-scratch | |
|---|---|---|---|
| Stars | 0 | — | 0 |
| Language | Python | Python | Python |
| Last pushed | — | 2022-11-22 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | easy | moderate | moderate |
| Complexity | 2/5 | 2/5 | 4/5 |
| Audience | developer | general | developer |
Figures from each repo's GitHub metadata at analysis time.
Vibes to Verified is a portable skill, meant to be installed into AI coding assistants like Claude Code, that pushes an AI agent to prove its work instead of just claiming it is done. The core idea is a five level scale that measures how much real evidence backs up a claim, running from V0, meaning a claim exists with nothing to support it, up to V4, meaning the finished result actually works in its real environment and someone can observe that it works. Alongside that scale, the skill tracks a separate verdict of green, yellow, red, blocked, or rejected, so a confident sounding answer with no proof and a carefully verified failure are never confused with each other. When applied to a task, the skill walks an agent through a defined process: it bounds what the agent is allowed to claim authority over, turns broad goals into specific claims that could be proven false, keeps track of more than one way to solve the problem, and then runs a dedicated round where independent reviewers try to poke holes in the final result. Failures have to be reproduced and checked again before they count as fixed, and everything ends in a machine readable evidence card that separates the final verdict from how mature the supporting evidence actually is. It installs as a project skill for Claude Code by cloning the repository into a skills folder, and it also supports the Hermes Agent platform along with any other agent framework that can read a SKILL.md file and its supporting reference and template documents. The project includes a validator script that checks whether an evidence card follows the required structure, though it cannot confirm on its own that an artifact is truthful or that a reviewer was genuinely independent. As a real world example, the author describes using the process while building an automated options paper trading risk control system, where independent reviews caught different classes of problems across several versions before the system was considered proven rather than just plausible.
A portable AI agent skill that grades claims on a five level evidence scale and forces independent verification before work counts as done.
Mainly Python. The stack also includes Python, Markdown, YAML.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.