fix(parser): prune qwenpaw source traversal

QwenPaw legacy discovery only walks valid workspace directories, the sessions directory, and one real non-hidden namespace below sessions. The provider migration reused the generic recursive JSONL walker, which preserved emitted source filtering but still allowed traversal and event classification through deeper or symlinked session namespaces.

Add a shared traversal predicate to the JSONL source helper so providers can keep discovery and changed-path classification aligned when their source layouts are narrower than an unbounded recursive scan.

Author: mariusvniekerk

internal/parser/jsonl_source_set.go

@@ -40,6 +40,10 @@ type JSONLSourceSetOptions struct {
 	// should enable it when legacy discovery accepted matching symlinked files
 	// and the parser reads through the symlink target.
 	FollowSymlinkFiles bool
+	// DescendPath is a directory predicate for recursive discovery. It is also
+	// applied to source ancestors during direct source classification so
+	// changed-path events cannot accept paths discovery would have pruned.
+	DescendPath func(root, path string) bool
 	// IncludePath is a path-only source predicate. It runs before Include and is
 	// also used for deleted/renamed changed paths where os.FileInfo is
 	// unavailable.
@@ -253,7 +257,7 @@ func (s JSONLSourceSet) discoverDir(
 		}
 		path := filepath.Join(dir, entry.Name())
 		if s.shouldDescend(entry, dir) {
-			if s.options.Recursive {
+			if s.options.Recursive && s.descendPathIncluded(root, path) {
 				if err := s.discoverDir(
 					ctx, root, path, sources, seen,
 				); err != nil {
@@ -292,6 +296,9 @@ func (s JSONLSourceSet) sourceForPath(path string) (SourceRef, bool) {
 		if !s.pathAllowedByRoot(root, path) {
 			continue
 		}
+		if !s.sourcePathAllowedByDescendPath(root, path) {
+			continue
+		}
 		if !s.pathIncluded(root, path) {
 			continue
 		}
@@ -324,6 +331,9 @@ func (s JSONLSourceSet) sourceForMissingPath(path string) (SourceRef, bool) {
 		if !s.pathAllowedByRoot(root, path) {
 			continue
 		}
+		if !s.sourcePathAllowedByDescendPath(root, path) {
+			continue
+		}
 		if !s.matchesExtension(path) || !s.pathIncluded(root, path) {
 			continue
 		}
@@ -411,6 +421,35 @@ func (s JSONLSourceSet) pathIncluded(root, path string) bool {
 	return s.options.IncludePath == nil || s.options.IncludePath(root, path)
 }
 
+func (s JSONLSourceSet) descendPathIncluded(root, path string) bool {
+	return s.options.DescendPath == nil || s.options.DescendPath(root, path)
+}
+
+func (s JSONLSourceSet) sourcePathAllowedByDescendPath(root, path string) bool {
+	if s.options.DescendPath == nil {
+		return true
+	}
+	rel, err := filepath.Rel(root, path)
+	if err != nil {
+		return false
+	}
+	dir := filepath.Dir(rel)
+	if dir == "." {
+		return true
+	}
+	current := root
+	for part := range strings.SplitSeq(dir, string(filepath.Separator)) {
+		if part == "" || part == "." || part == ".." {
+			return false
+		}
+		current = filepath.Join(current, part)
+		if !s.descendPathIncluded(root, current) {
+			return false
+		}
+	}
+	return true
+}
+
 func (s JSONLSourceSet) matchesExtension(path string) bool {
 	ext := filepath.Ext(path)
 	return slices.Contains(s.extensions, ext)

internal/parser/jsonl_source_set_test.go

@@ -266,6 +266,44 @@ func TestJSONLSourceSetChangedPathUsesPathOnlyFilterForDeletedFiles(t *testing.T
 	assert.Equal(t, filepath.Join(root, "session", "events.jsonl"), changed[0].DisplayPath)
 }
 
+func TestJSONLSourceSetDescendPathPrunesSources(t *testing.T) {
+	root := t.TempDir()
+	keepPath := filepath.Join(root, "keep", "session.jsonl")
+	skipPath := filepath.Join(root, "skip", "session.jsonl")
+	writeSourceFile(t, keepPath, "{}\n")
+	writeSourceFile(t, skipPath, "{}\n")
+
+	sources := NewJSONLSourceSet(AgentCodex, []string{root}, JSONLSourceSetOptions{
+		Recursive: true,
+		DescendPath: func(root, path string) bool {
+			return filepath.Base(path) != "skip"
+		},
+	})
+
+	discovered, err := sources.Discover(context.Background())
+	require.NoError(t, err)
+	require.Len(t, discovered, 1)
+	assert.Equal(t, keepPath, discovered[0].DisplayPath)
+
+	changed, err := sources.SourcesForChangedPath(
+		context.Background(),
+		ChangedPathRequest{Path: skipPath, EventKind: "write", WatchRoot: root},
+	)
+	require.NoError(t, err)
+	assert.Empty(t, changed)
+
+	removed, err := sources.SourcesForChangedPath(
+		context.Background(),
+		ChangedPathRequest{
+			Path:      filepath.Join(root, "skip", "removed.jsonl"),
+			EventKind: "remove",
+			WatchRoot: root,
+		},
+	)
+	require.NoError(t, err)
+	assert.Empty(t, removed)
+}
+
 func TestJSONLSourceSetDuplicateKeysKeepFirstConfiguredRoot(t *testing.T) {
 	firstRoot := t.TempDir()
 	secondRoot := t.TempDir()

internal/parser/qwenpaw_provider.go

@@ -3,6 +3,7 @@ package parser
 import (
 	"context"
 	"fmt"
+	"os"
 	"path/filepath"
 	"strings"
 )
@@ -140,6 +141,7 @@ func newQwenPawSourceSet(roots []string) JSONLSourceSet {
 		Hash:               true,
 		FollowSymlinkDirs:  true,
 		FollowSymlinkFiles: true,
+		DescendPath:        qwenPawDescendPath,
 		IncludePath:        isQwenPawSourcePath,
 		ProjectHint:        qwenPawProjectHintFromPath,
 		SessionIDFromPath:  qwenPawSessionIDFromPath,
@@ -207,6 +209,34 @@ func qwenPawSourcePathParts(root, path string) ([]string, bool) {
 	return parts, true
 }
 
+func qwenPawDescendPath(root, path string) bool {
+	parts, ok := qwenPawSourcePathParts(root, path)
+	if !ok {
+		return false
+	}
+	switch len(parts) {
+	case 1:
+		return IsValidQwenPawIDPart(parts[0])
+	case 2:
+		return IsValidQwenPawIDPart(parts[0]) && parts[1] == "sessions"
+	case 3:
+		subdir := parts[2]
+		if parts[1] != "sessions" ||
+			!IsValidQwenPawIDPart(parts[0]) ||
+			strings.HasPrefix(subdir, ".") ||
+			!IsValidQwenPawIDPart(subdir) {
+			return false
+		}
+		info, err := os.Lstat(path)
+		if err != nil {
+			return true
+		}
+		return info.Mode()&os.ModeSymlink == 0
+	default:
+		return false
+	}
+}
+
 func qwenPawProviderCapabilities() Capabilities {
 	source := jsonlFileProviderSourceCapabilities()
 	source.ForceReplaceOnParse = CapabilitySupported

internal/parser/qwenpaw_provider_test.go

@@ -189,6 +189,47 @@ func TestQwenPawProviderDiscoversSymlinkedWorkspace(t *testing.T) {
 	assert.Equal(t, "default", discovered[0].ProjectHint)
 }
 
+func TestQwenPawProviderPrunesSymlinkedSessionNamespaces(t *testing.T) {
+	root := t.TempDir()
+	sourcePath := qwenPawProviderWriteSession(
+		t, root, "default", "", "root_1", "root question",
+	)
+	targetDir := filepath.Join(t.TempDir(), "console-target")
+	require.NoError(t, os.MkdirAll(targetDir, 0o755))
+	require.NoError(t, os.WriteFile(
+		filepath.Join(targetDir, "linked_1.json"),
+		[]byte(qwenPawProviderFixture("linked question")),
+		0o644,
+	))
+	linkedDir := filepath.Join(root, "default", "sessions", "linked")
+	if err := os.Symlink(targetDir, linkedDir); err != nil {
+		t.Skipf("symlink not supported: %v", err)
+	}
+	linkedPath := filepath.Join(linkedDir, "linked_1.json")
+
+	provider, ok := NewProvider(AgentQwenPaw, ProviderConfig{
+		Roots:   []string{root},
+		Machine: "devbox",
+	})
+	require.True(t, ok)
+
+	discovered, err := provider.Discover(context.Background())
+	require.NoError(t, err)
+	require.Len(t, discovered, 1)
+	assert.Equal(t, sourcePath, discovered[0].DisplayPath)
+
+	changed, err := provider.SourcesForChangedPath(
+		context.Background(),
+		ChangedPathRequest{
+			Path:      linkedPath,
+			EventKind: "write",
+			WatchRoot: root,
+		},
+	)
+	require.NoError(t, err)
+	assert.Empty(t, changed)
+}
+
 func qwenPawProviderWriteSession(
 	t *testing.T,
 	root string,