Driftproofhq

Open-source receipts for AI agent skills: does a skill still help after a model update? Measured per release, with the evidence kept. Paper: doi.org/10.5281/zenodo.23050795

Joined on Sep 20, 2026 https://driftproofhq.com

Driftproofhq

Oct 9

We Checked 162 AI Benchmark Gaps. Only 20 Separate Cleanly. The Bigger Problem Was the Missing Data.

#ai #machinelearning #llm #opensource

4 min read

Want to connect with Driftproofhq?

Create an account to connect with Driftproofhq. You can also sign in below to proceed if you already have an account.

Create Account

Already have an account? Sign in

Driftproofhq

Oct 8

I ran the same Claude Code skill on Haiku 5.5 twice. Run 1: +55 points. Run 2: nothing. Driftproofhq

#ai #claude #llm #testing

2 min read

Driftproofhq

Oct 6

A well-formed number is not a measurement: 13 defects across 7 eval tools

#machinelearning #testing #opensource #ai

3 min read

Driftproofhq

Sep 29

The same git skill added almost nothing to Sonnet 5.5 and took Opus 5.5 from about 0.5 to 0.86

#ai #claude #llm #testing

4 min read

Driftproofhq

Sep 24

Opus 5.5 shipped at 16:31 UTC. Within 15 hours, we had receipts on whether three agent skills still help.

#ai #llm #testing #opensource

3 min read

Driftproofhq

Sep 20

The biggest improvement in my skill evaluation came from a skill that was never invoked

#ai #llm #testing #opensource

4 min read

DEV Community

Driftproofhq

Badges

1 Week Community Wellness Streak

Writing Debut

We Checked 162 AI Benchmark Gaps. Only 20 Separate Cleanly. The Bigger Problem Was the Missing Data.

Want to connect with Driftproofhq?

I ran the same Claude Code skill on Haiku 5.5 twice. Run 1: +55 points. Run 2: nothing. Driftproofhq

A well-formed number is not a measurement: 13 defects across 7 eval tools

The same git skill added almost nothing to Sonnet 5.5 and took Opus 5.5 from about 0.5 to 0.86

Opus 5.5 shipped at 16:31 UTC. Within 15 hours, we had receipts on whether three agent skills still help.

The biggest improvement in my skill evaluation came from a skill that was never invoked