From 5e5be4b86df4f2f36eb76636bcc4dcecc11a16f7 Mon Sep 17 00:00:00 2001 From: NhoNH Date: Fri, 5 Jun 2026 13:20:04 +0000 Subject: [PATCH] docs: add eval-harness to projects --- data/projects/eval-harness.yaml | 4 ++++ 1 file changed, 4 insertions(+) create mode 100644 data/projects/eval-harness.yaml diff --git a/data/projects/eval-harness.yaml b/data/projects/eval-harness.yaml new file mode 100644 index 0000000..0a5a33c --- /dev/null +++ b/data/projects/eval-harness.yaml @@ -0,0 +1,4 @@ +name: eval-harness +repo: https://github.com/nano-step/eval-harness +tagline: Behavior-regression testing for OpenCode skills +description: Detects when an OpenCode skill's behavior drifts from a baseline, attributes the cause across 4 classes (skill change / fixture stale / model change / unknown drift), and emits a 6-field FAIL schema with transcript-span and env-delta evidence. Ships a git pre-push hook, a composite GitHub Action, and a $-cost hard ceiling for CI safety. Bash + jq + python3 stdlib only; no daemon, no SaaS. MIT.