diff --git a/evals/harbor/.agents/skills/compare_tasks/SKILL.md b/evals/harbor/.agents/skills/compare-tasks/SKILL.md
similarity index 99%
rename from evals/harbor/.agents/skills/compare_tasks/SKILL.md
rename to evals/harbor/.agents/skills/compare-tasks/SKILL.md
index 49877caf7..68e24e266 100644
--- a/evals/harbor/.agents/skills/compare_tasks/SKILL.md
+++ b/evals/harbor/.agents/skills/compare-tasks/SKILL.md
@@ -1,5 +1,5 @@
 ---
-name: compare_tasks
+name: compare-tasks
 description: Compare how two harbor benchmark runs performed on a single shared task
 ---
 
diff --git a/evals/harbor/README.md b/evals/harbor/README.md
index e307fc2ff..3068aac39 100644
--- a/evals/harbor/README.md
+++ b/evals/harbor/README.md
@@ -231,7 +231,7 @@ cd /home/tbench/work/goose
 ```
 
 For deeper per-task understanding (why did A pass and B fail on this one
-task?), see the `compare_tasks` skill under `.agents/skills/`. Delegate to
+task?), see the `compare-tasks` skill under `.agents/skills/`. Delegate to
 it with the two job names and a task name and it will read both
 trajectories, the task spec, and the verifier output, then explain the
 mechanism behind the divergence.