Harden skill-eval + A/B-polish MAP agents (−1049 lines of example bloat)#161

Merged

azalio merged 24 commits into

mainfrom

Jun 6, 2026

Commits on Jun 5, 2026

fix(skill-eval): trigger-eval no longer mis-records executing skills as non-trigger

azalio
and
claude
committed
fix(skill-eval): optimizer corrupted block-scalar SKILL.md descriptions → skill never triggered

azalio
and
claude
committed
feat(skill-eval): --model/--runs flags + model-tier trigger experiment

azalio
and
claude
committed
docs(skill-eval): firm-up model-tier results revise the pilot conclusion

azalio
and
claude
committed
feat(whole-skill): model-lever knobs + a discriminating outcome fixture

azalio
and
claude
committed
docs(whole-skill): model lever on OUTCOME is absorbed by the gate (semver fixture)

azalio
and
claude
committed
feat(whole-skill): capture serial retry_count in the outcome harness

azalio
and
claude
committed
feat(whole-skill): weakly-gated fixture + hidden-suite scoring (model-effect probe)

azalio
and
claude
committed
test(whole-skill): vague-contract fixture to isolate the model lever

azalio
and
claude
committed
docs(whole-skill): outcome is invariant to model/gate/contract within competence

azalio
and
claude
committed
test(whole-skill): capability-discriminating calc fixture (expression evaluator)

azalio
and
claude
committed
test(whole-skill): vague-contract calc fixture (decisive model-threshold probe)

azalio
and
claude
committed
docs(whole-skill): capability probe — haiku matches bigger models on subtasks

azalio
and
claude
committed
fix(whole-skill): acceptEdits in outcome harness + use orchestrator-model

azalio
and
claude
committed

Commits on Jun 6, 2026