M5: v0.5 evals - #13

Merged

hazeliscoding merged 43 commits into

mainfrom

Sep 30, 2026

Commits on Sep 29, 2026

docs(roadmap): plan M5, evals
hazeliscoding
committed
812b739
test(evals): add the M5 spike's harness streams as fixtures
hazeliscoding
committed
b358914
docs(roadmap): record what the M5 spike settled, and seal Codex by disabling the real home's skills
hazeliscoding
committed
bf0991c
feat(schemas): add eval cases, checked by the doctor
hazeliscoding
committed
7549a0f
feat(evals): decide an eval case's checks from what the harness recorded
hazeliscoding
committed
431f1ee
feat(evals): read a whole session from each harness, subagents included
hazeliscoding
committed
320d870
docs(roadmap): run M5's evals sealed only, and move --real-home to Later
hazeliscoding
committed
4231a42
feat(evals): install an asset for each harness the way sync will
hazeliscoding
committed
74b1170
feat(evals): plan a sealed home that borrows only the logins and model choices
hazeliscoding
committed
a215ad8
feat(evals): plan an eval run from the vault's cases
hazeliscoding
committed
835f6af

Commits on Sep 30, 2026