fix: hidden test exposure & docs: bump version to v1.6.0 with incompatibility note

Author: lemon07r

README.md

@@ -9,6 +9,8 @@
 
 A lightweight evaluation harness for coding agents that runs high-signal, compact but challenging problems in isolated Docker containers. Evaluate agents across 26 tasks in 6 languages with weighted scoring, integrity verification, and detailed reporting.
 
+> **Note:** All evaluation results obtained before version `v1.6.0` cannot be compared to results obtained on or after `v1.6.0` due to a critical fix in how hidden tests are handled.
+
 <!-- Add demo GIF/screenshot here -->
 
 ## Table of Contents

internal/runner/runner.go

@@ -412,7 +412,7 @@ func (r *Runner) ensureWorkspace(t *task.Task, dir string) error {
 		return nil
 	}
 
-	return r.copyTaskFiles(t, dir, t.AllFiles())
+	return r.copyTaskFiles(t, dir, t.VisibleFiles())
 }
 
 // captureWorkspace reads the workspace files into the session.
@@ -500,7 +500,7 @@ func (r *Runner) InitWorkspaceForTask(t *task.Task, outputDir string) error {
 		return fmt.Errorf("directory is not empty: %s", absDir)
 	}
 
-	return r.copyTaskFiles(t, absDir, t.AllFiles())
+	return r.copyTaskFiles(t, absDir, t.VisibleFiles())
 }
 
 // ListTasks returns all available tasks.

internal/task/task.go

@@ -69,6 +69,15 @@ type Validation struct {
 	Args    []string `json:"args"    toml:"args"`
 }
 
+// VisibleFiles returns the files that should be visible to the agent initially.
+func (t *Task) VisibleFiles() []string {
+	files := make([]string, 0, len(t.Files.Stub)+len(t.Files.Test)+len(t.Files.Support))
+	files = append(files, t.Files.Stub...)
+	files = append(files, t.Files.Test...)
+	files = append(files, t.Files.Support...)
+	return files
+}
+
 // AllFiles returns all files associated with this task, including hidden tests.
 func (t *Task) AllFiles() []string {
 	files := make([]string, 0, len(t.Files.Stub)+len(t.Files.Test)+len(t.Files.HiddenTest)+len(t.Files.Support))