Opensiro
500+ agent harness assessments #18
xLagerFeuer
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
500+ agent harness assessments in the VSM Harness Index!
Main updates since our 250-assessment milestone:
The Index keeps growing. We now have over 600 evidence-backed assessments, mapping how different agent systems implement S1-S5 organizational functions, with pinned source references and explicit autonomy boundaries.
Real upstream improvements. Our work has already contributed to merged changes in Marveen and Henterprise, covering owner-gated skill adoption, task-scoped coordination, and independent review.
Building harnesses using the Index. We've started using assessments as an architectural knowledge base. Our Cognee-first experiments explore how existing harness implementations can inform the design and construction of new systems.
Beyond autonomy classification. We're also developing experimental methods to compare the actual capabilities of individual organizational functions using empirical evidence, separately from their VSM autonomy states.
What we're working on now:
We're developing a repeatable process for improving open-source agent systems:
VSM assessment → functional gap → product-level validation → implementation → before/after evaluation → upstream contribution.
The next challenge is automating this workflow and scaling it toward improvements across 100 open-source projects.
The goal is not simply to produce more assessments or pull requests, but to make the accumulated evidence useful for building and improving real agent systems.
All reactions