From d33f1b3c9d955954aba65948de599701b52b7061 Mon Sep 17 00:00:00 2001
From: anmarhindi <153752112+anmarhindi@users.noreply.github.com>
Date: Thu, 14 May 2026 17:03:08 +0200
Subject: [PATCH 1/2] fix: suppress misleading "Did you mean" suggestions

---
 pkg/cmd/suggest.go      | 19 +++++++++--
 pkg/cmd/suggest_test.go | 70 +++++++++++++++++++++++++++++++++++++++++
 2 files changed, 87 insertions(+), 2 deletions(-)
 create mode 100644 pkg/cmd/suggest_test.go

diff --git a/pkg/cmd/suggest.go b/pkg/cmd/suggest.go
index b4b637c0..a7efa3e6 100644
--- a/pkg/cmd/suggest.go
+++ b/pkg/cmd/suggest.go
@@ -98,10 +98,22 @@ func jaroWinkler(a, b string) float64 {
 	return jaroDist + 0.1*prefixMatch*(1.0-jaroDist)
 }
 
+// suggestionThreshold is the minimum jaro-winkler similarity required before we
+// will print a "Did you mean" suggestion. Below this, the closest match is too
+// dissimilar to be a useful guess, so we stay silent rather than mislead.
+// 0.7 matches the boostThreshold used by jaroWinkler above — the prefix boost
+// only kicks in past that, so it's a natural "plausibly the same word" cutoff.
+const suggestionThreshold = 0.7
+
 // suggestCommand takes a list of commands and a provided string to suggest a
-// command name
+// command name. Returns an empty string when no command is sufficiently
+// similar; the upstream urfave/cli error formatter omits the suggestion clause
+// in that case.
 func suggestCommand(commands []*cli.Command, provided string) string {
-	distance := 0.0
+	if provided == "" {
+		return ""
+	}
+	distance := suggestionThreshold
 	var lineage []*cli.Command
 	for _, command := range commands {
 		for _, name := range command.Names() {
@@ -112,6 +124,9 @@ func suggestCommand(commands []*cli.Command, provided string) string {
 			}
 		}
 	}
+	if lineage == nil {
+		return ""
+	}
 
 	var parts []string
 	for _, command := range lineage {
diff --git a/pkg/cmd/suggest_test.go b/pkg/cmd/suggest_test.go
new file mode 100644
index 00000000..43c79d27
--- /dev/null
+++ b/pkg/cmd/suggest_test.go
@@ -0,0 +1,70 @@
+package cmd
+
+import (
+	"testing"
+
+	"github.com/urfave/cli/v3"
+)
+
+func TestSuggestCommand(t *testing.T) {
+	commands := []*cli.Command{
+		{Name: "create"},
+		{Name: "retrieve"},
+		{Name: "list"},
+		{Name: "delete"},
+		{Name: "chat:completions"},
+		{Name: "completions"},
+	}
+
+	tests := []struct {
+		name     string
+		provided string
+		want     string
+	}{
+		{
+			name:     "close typo suggests the corrected command",
+			provided: "creat",
+			want:     "Did you mean 'create'?",
+		},
+		{
+			name:     "near-exact suggests the corrected command",
+			provided: "chat:completion",
+			want:     "Did you mean 'chat:completions'?",
+		},
+		{
+			name:     "exact match still suggests itself",
+			provided: "create",
+			want:     "Did you mean 'create'?",
+		},
+		{
+			name:     "unrelated input returns no suggestion",
+			provided: "zzzzz",
+			want:     "",
+		},
+		{
+			name:     "low-similarity input returns no suggestion",
+			provided: "totallybogus",
+			want:     "",
+		},
+		{
+			name:     "empty input returns no suggestion",
+			provided: "",
+			want:     "",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got := suggestCommand(commands, tc.provided)
+			if got != tc.want {
+				t.Errorf("suggestCommand(%q) = %q, want %q", tc.provided, got, tc.want)
+			}
+		})
+	}
+}
+
+func TestSuggestCommandEmptyCommands(t *testing.T) {
+	if got := suggestCommand(nil, "anything"); got != "" {
+		t.Errorf("suggestCommand(nil, %q) = %q, want empty string", "anything", got)
+	}
+}

From 8e1d3c7de2bd4f8c45e593d774bb6e75890c9fe6 Mon Sep 17 00:00:00 2001
From: anmarhindi <153752112+anmarhindi@users.noreply.github.com>
Date: Mon, 21 Sep 2026 15:59:01 +0200
Subject: [PATCH 2/2] fix: keep short-typo and case-insensitive command
 suggestions

Names within one edit of the input (insertion, deletion, substitution, or
adjacent transposition) are now always suggested and outrank other names,
so `rn`/`rnu` suggest `run` again and long names sharing a prefix are no
longer misranked. Scoring ignores case, so `RESPONSES` suggests
`responses`. The 0.7 threshold tolerates float error, since an exact 7/10
jaro score computes as 0.7000000000000001.

Adds tests against the real command tree and an end-to-end check through
the entrypoint.
---
 cmd/openai/main_suggest_test.go |  29 +++++++
 pkg/cmd/suggest.go              |  44 +++++++++--
 pkg/cmd/suggest_test.go         | 135 ++++++++++++++++++++++++++++++++
 3 files changed, 203 insertions(+), 5 deletions(-)
 create mode 100644 cmd/openai/main_suggest_test.go

diff --git a/cmd/openai/main_suggest_test.go b/cmd/openai/main_suggest_test.go
new file mode 100644
index 00000000..35cbb3a2
--- /dev/null
+++ b/cmd/openai/main_suggest_test.go
@@ -0,0 +1,29 @@
+package main
+
+import (
+	"strings"
+	"testing"
+)
+
+// Unknown commands suggest the full path of a close match and stay silent
+// when nothing is close.
+func TestMainUnknownCommandSuggestions(t *testing.T) {
+	for _, tc := range []struct {
+		args   []string
+		stderr string
+	}{
+		{[]string{"fine-tuning:alpha:graders", "rn"}, "No help topic for 'rn'. Did you mean 'openai fine-tuning:alpha:graders run'?\n"},
+		{[]string{"fine-tuning:alpha:graders", "rnu"}, "No help topic for 'rnu'. Did you mean 'openai fine-tuning:alpha:graders run'?\n"},
+		{[]string{"responses", "creat"}, "No help topic for 'creat'. Did you mean 'openai responses create'?\n"},
+		{[]string{"RESPONSES"}, "No help topic for 'RESPONSES'. Did you mean 'openai responses'?\n"},
+		{[]string{"totallybogus"}, "No help topic for 'totallybogus'\n"},
+		{[]string{"responses", "zzzzz"}, "No help topic for 'zzzzz'\n"},
+	} {
+		t.Run(strings.Join(tc.args, "/"), func(t *testing.T) {
+			want := mainDispatchResult{3, "", tc.stderr}
+			if got := runMainDispatch(t, "bash", append([]string{"openai"}, tc.args...)...); got != want {
+				t.Fatalf("got %+v; want %+v", got, want)
+			}
+		})
+	}
+}
diff --git a/pkg/cmd/suggest.go b/pkg/cmd/suggest.go
index a7efa3e6..a2af6859 100644
--- a/pkg/cmd/suggest.go
+++ b/pkg/cmd/suggest.go
@@ -10,7 +10,7 @@ import (
 )
 
 // This entire file is mostly taken from urfave/cli/v3's source, with the exception of suggestCommand which is
-// modified for a nicer error message.
+// modified for a nicer error message and stricter matching, and its helpers suggestionThreshold and withinOneEdit.
 
 // jaroDistance is the measure of similarity between two strings. It returns a
 // value between 0 and 1, where 1 indicates identical strings and 0 indicates
@@ -103,21 +103,55 @@ func jaroWinkler(a, b string) float64 {
 // dissimilar to be a useful guess, so we stay silent rather than mislead.
 // 0.7 matches the boostThreshold used by jaroWinkler above — the prefix boost
 // only kicks in past that, so it's a natural "plausibly the same word" cutoff.
+// Names within one edit of the input are exempt; see suggestCommand.
 const suggestionThreshold = 0.7
 
+// withinOneEdit reports whether a and b differ by at most one insertion,
+// deletion, substitution, or transposition of adjacent characters.
+func withinOneEdit(a, b string) bool {
+	if len(a) < len(b) {
+		a, b = b, a
+	}
+	if len(a)-len(b) > 1 {
+		return false
+	}
+	i := 0
+	for i < len(b) && a[i] == b[i] {
+		i++
+	}
+	if i == len(a) {
+		return true
+	}
+	if len(a) != len(b) {
+		return a[i+1:] == b[i:]
+	}
+	return a[i+1:] == b[i+1:] ||
+		(i+1 < len(a) && a[i] == b[i+1] && a[i+1] == b[i] && a[i+2:] == b[i+2:])
+}
+
 // suggestCommand takes a list of commands and a provided string to suggest a
-// command name. Returns an empty string when no command is sufficiently
-// similar; the upstream urfave/cli error formatter omits the suggestion clause
-// in that case.
+// command name, ignoring case. A name within one edit of the input is always
+// suggested and outranks every other name, because jaro-winkler undervalues
+// typos in short names ("rn" for "run") and misranks long names that share a
+// prefix. Otherwise the closest name must exceed suggestionThreshold. Returns an
+// empty string when no command is sufficiently similar; the upstream urfave/cli
+// error formatter omits the suggestion clause in that case.
 func suggestCommand(commands []*cli.Command, provided string) string {
+	provided = strings.ToLower(provided)
 	if provided == "" {
 		return ""
 	}
-	distance := suggestionThreshold
+	// An exact 7/10 jaro score computes as 0.7000000000000001, so allow for
+	// float error before counting the threshold as passed.
+	distance := suggestionThreshold + 1e-9
 	var lineage []*cli.Command
 	for _, command := range commands {
 		for _, name := range command.Names() {
+			name = strings.ToLower(name)
 			newDistance := jaroWinkler(name, provided)
+			if withinOneEdit(name, provided) {
+				newDistance++ // outranks every name further away
+			}
 			if newDistance > distance {
 				distance = newDistance
 				lineage = command.Lineage()
diff --git a/pkg/cmd/suggest_test.go b/pkg/cmd/suggest_test.go
index 43c79d27..4ee51537 100644
--- a/pkg/cmd/suggest_test.go
+++ b/pkg/cmd/suggest_test.go
@@ -1,6 +1,8 @@
 package cmd
 
 import (
+	"fmt"
+	"strings"
 	"testing"
 
 	"github.com/urfave/cli/v3"
@@ -36,6 +38,11 @@ func TestSuggestCommand(t *testing.T) {
 			provided: "create",
 			want:     "Did you mean 'create'?",
 		},
+		{
+			name:     "uppercase input is matched ignoring case",
+			provided: "CREAT",
+			want:     "Did you mean 'create'?",
+		},
 		{
 			name:     "unrelated input returns no suggestion",
 			provided: "zzzzz",
@@ -68,3 +75,131 @@ func TestSuggestCommandEmptyCommands(t *testing.T) {
 		t.Errorf("suggestCommand(nil, %q) = %q, want empty string", "anything", got)
 	}
 }
+
+func TestWithinOneEdit(t *testing.T) {
+	tests := []struct {
+		a, b string
+		want bool
+	}{
+		{"", "", true},
+		{"", "a", true},
+		{"", "ab", false},
+		{"run", "run", true},
+		{"run", "rn", true},   // deletion
+		{"run", "runn", true}, // insertion
+		{"run", "ran", true},  // substitution
+		{"run", "rnu", true},  // transposition
+		{"run", "urn", true},  // transposition
+		{"ab", "ba", true},    // transposition
+		{"run", "rm", false},
+		{"run", "nur", false},
+		{"list", "ls", false},
+		{"create", "craete", true},
+		{"create", "carete", false},
+		{"create", "caerte", false},
+		{"run", "urm", false}, // swapped pair plus another change
+		{"run", "xrn", false}, // only one side of the swap matches
+	}
+	for _, tc := range tests {
+		for _, pair := range [][2]string{{tc.a, tc.b}, {tc.b, tc.a}} {
+			if got := withinOneEdit(pair[0], pair[1]); got != tc.want {
+				t.Errorf("withinOneEdit(%q, %q) = %v, want %v", pair[0], pair[1], got, tc.want)
+			}
+		}
+	}
+}
+
+// findCommand resolves a command path in the real command tree.
+func findCommand(t *testing.T, path ...string) *cli.Command {
+	t.Helper()
+	command := Command
+	for _, name := range path {
+		var next *cli.Command
+		for _, sub := range command.Commands {
+			if sub.Name == name {
+				next = sub
+			}
+		}
+		if next == nil {
+			t.Fatalf("command %q not found under %q", name, command.Name)
+		}
+		command = next
+	}
+	return command
+}
+
+// Parent links are only set once the command tree runs, so these suggestions
+// name the matched command alone; cmd/openai checks the full "openai ..." form.
+func TestSuggestCommandRealTree(t *testing.T) {
+	tests := []struct {
+		path     []string
+		provided string
+		want     string
+	}{
+		{[]string{"fine-tuning:alpha:graders"}, "rn", "run"},
+		{[]string{"fine-tuning:alpha:graders"}, "rnu", "run"},
+		{[]string{"fine-tuning:alpha:graders"}, "urn", "run"},
+		{[]string{"fine-tuning:alpha:graders"}, "RN", "run"},
+		{[]string{"fine-tuning:alpha:graders"}, "valdate", "validate"},
+		{[]string{"fine-tuning:alpha:graders"}, "zzz", ""},
+		{[]string{"responses"}, "zzzzz", ""},
+		{[]string{"models"}, "ls", "list"},     // jaro-winkler 0.85
+		{[]string{"models"}, "lete", "delete"}, // jaro-winkler 0.72
+		// Both are one edit away; the higher jaro-winkler score wins.
+		{[]string{"admin:organization:certificates"}, "dactivate", "deactivate"},
+		// Exact 7/10 jaro scores that float error would otherwise let through.
+		{[]string{"fine-tuning:jobs"}, "status", ""},
+		{nil, "hi", ""},
+		{nil, "RESPONSES", "responses"},
+		{nil, "respones", "responses"},
+		{nil, "chat:completion", "chat:completions"},
+		{nil, "comp", "completions"},
+		// Jaro-winkler alone prefers admin:organization:spend-alerts here.
+		{nil, "admin:prganization:roles", "admin:organization:roles"},
+		{nil, "version", ""}, // jaro-winkler 0.69
+		{nil, "totallybogus", ""},
+	}
+	for _, tc := range tests {
+		t.Run(strings.Join(append(tc.path, tc.provided), " "), func(t *testing.T) {
+			want := ""
+			if tc.want != "" {
+				want = fmt.Sprintf("Did you mean '%s'?", tc.want)
+			}
+			if got := suggestCommand(findCommand(t, tc.path...).Commands, tc.provided); got != want {
+				t.Errorf("got %q, want %q", got, want)
+			}
+		})
+	}
+}
+
+// Every command in the real tree keeps its suggestion for a dropped or swapped
+// character or for all-caps input, unless the typo is as close to a sibling.
+func TestSuggestCommandRealTreeSingleEdits(t *testing.T) {
+	var visit func(parent *cli.Command)
+	visit = func(parent *cli.Command) {
+		for _, command := range parent.Commands {
+			name := command.Name
+			typos := []string{strings.ToUpper(name)}
+			for i := range name {
+				typos = append(typos, name[:i]+name[i+1:])
+				if i+1 < len(name) {
+					typos = append(typos, name[:i]+name[i+1:i+2]+name[i:i+1]+name[i+2:])
+				}
+			}
+		typos:
+			for _, typo := range typos {
+				for _, sibling := range parent.Commands {
+					if sibling != command && withinOneEdit(sibling.Name, strings.ToLower(typo)) {
+						continue typos
+					}
+				}
+				want := fmt.Sprintf("Did you mean '%s'?", name)
+				if got := suggestCommand(parent.Commands, typo); got != want {
+					t.Errorf("%s: suggestCommand(%q) = %q, want %q", parent.Name, typo, got, want)
+				}
+			}
+			visit(command)
+		}
+	}
+	visit(Command)
+}