fix: fail diff when skills exceed absolute token limits, improve wording

spboyer · Copilot · spboyer · commit 4208dd689bec · 2026-03-09T18:02:15.000-04:00
- Passed flag now checks both ThresholdExceeded and OverLimit — a skill over its absolute budget causes failure regardless of percent-change threshold - Clarify Long description with fallback behavior - Add inline comment above base ref fallback - Generalize error message from 'threshold exceeded' to 'budget exceeded' Addresses wbreza [High] and chlowell review feedback on PR #93. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
diff --git a/cmd/waza/tokens/diff.go b/cmd/waza/tokens/diff.go
@@ -28,9 +28,9 @@ func newDiffCmd() *cobra.Command {
 		Short: "Compare SKILL.md token budgets against a base ref",
 		Long: `Compare token budgets for SKILL.md files between a base git ref and current changes.
 
-By default, compares origin/main to the working tree and reports per-skill deltas.
-Skill roots include configured paths.skills from .waza.yaml plus skills/ and
-.github/skills/.`,
+By default the base ref is origin/main, falling back to main when origin/main
+is not available. Skill roots include configured paths.skills from .waza.yaml
+plus skills/ and .github/skills/.`,
 		Args:          cobra.MaximumNArgs(1),
 		RunE:          runDiff,
 		SilenceErrors: true,
@@ -92,6 +92,7 @@ func runDiff(cmd *cobra.Command, args []string) error {
 		return errors.New("not a git repository; diff command requires git")
 	}
 
+	// Prefers origin/main, falls back to main.
 	baseRef := defaultDiffBaseRef
 	if len(args) == 1 {
 		baseRef = args[0]
@@ -110,7 +111,7 @@ func runDiff(cmd *cobra.Command, args []string) error {
 	}
 	passed := true
 	for _, d := range diffs {
-		if d.ThresholdExceeded {
+		if d.ThresholdExceeded || d.OverLimit {
 			passed = false
 			break
 		}
@@ -142,7 +143,7 @@ func runDiff(cmd *cobra.Command, args []string) error {
 
 	if !passed {
 		cmd.SilenceUsage = true
-		return fmt.Errorf("token diff threshold exceeded for one or more skills (threshold %.1f%%)", threshold)
+		return fmt.Errorf("token budget exceeded for one or more skills")
 	}
 	return nil
 }
diff --git a/cmd/waza/tokens/diff_test.go b/cmd/waza/tokens/diff_test.go
@@ -38,7 +38,7 @@ func TestDiff_ThresholdAndSkillRoots(t *testing.T) {
 
 	err := cmd.Execute()
 	require.Error(t, err)
-	require.Contains(t, err.Error(), "threshold exceeded")
+	require.Contains(t, err.Error(), "budget exceeded")
 	require.Contains(t, out.String(), "skills/alpha")
 	require.Contains(t, out.String(), ".github/skills/beta")
 	require.Contains(t, out.String(), "⚠️")
@@ -68,13 +68,14 @@ tokens:
 	cmd.SetErr(new(bytes.Buffer))
 	cmd.SetArgs([]string{"main", "--format", "json", "--threshold", "500"})
 
-	require.NoError(t, cmd.Execute())
+	err := cmd.Execute()
+	require.Error(t, err, "should fail because skill is over its absolute token limit")
 
 	var report diffReport
 	require.NoError(t, json.Unmarshal(out.Bytes(), &report))
 	require.Equal(t, "main", report.BaseRef)
 	require.Equal(t, "WORKING", report.HeadRef)
-	require.True(t, report.Passed)
+	require.False(t, report.Passed, "Passed should be false when a skill is over its absolute limit")
 	require.Len(t, report.Skills, 1)
 	require.Equal(t, "custom-skills/delta", report.Skills[0].Path)
 	require.Equal(t, 5, report.Skills[0].Limit)
