fix(linkcheck): skip a linked tree that is not a real directory (#157)

Review round 1 on PR #175:

- entryIndex returned WalkRegular's walk-root ErrNotRegular as an error,
  and StripUnresolvedLinks runs before every brain_write: a symlinked,
  dangling or regular-file knowledge/ (or wiki/) made every write fail
  and lose the learning. main's filepath.WalkDir skipped such a root. It
  is now skipped with a warning; other walk errors still abort, as on
  main. New table test over the three shapes.
- The symlinked-hall assertion passed with no index written at all. It
  now indexes a real hall next to the symlinked one and asserts no
  error, the index exists and lists the real note; a new test does the
  same for BuildAllWingIndexes with a symlinked wing next to a real one.
- The test comment no longer says this PR fixed the FIFO hang on every
  brain_write: that path is tunnel.DetectTunnels (slice 5b).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

ingestion/cmd/server/jwt_validator.go

@@ -0,0 +1,25 @@
+package main
+
+import (
+	"context"
+	"log/slog"
+
+	chassisauth "git.d-ma.be/mathias/mcp-chassis/auth"
+
+	"github.com/mathiasbq/hyperguild/ingestion/internal/startupretry"
+)
+
+// buildJWTValidator runs OIDC discovery under policy, so a 502 from an auth
+// upstream that is not serving yet on a cold boot is retried instead of
+// exiting (#83). It honours ctx between attempts.
+func buildJWTValidator(
+	ctx context.Context, logger *slog.Logger, issuer, audience string, policy startupretry.Policy,
+) (*chassisauth.JWTValidator, error) {
+	return startupretry.Do(ctx, policy, func(ctx context.Context) (*chassisauth.JWTValidator, error) {
+		v, err := chassisauth.NewJWTValidator(ctx, issuer, audience)
+		if err != nil {
+			logger.Warn("oidc discovery not ready, retrying", "err", err)
+		}
+		return v, err
+	})
+}

ingestion/cmd/server/jwt_validator_test.go

@@ -0,0 +1,86 @@
+package main
+
+import (
+	"context"
+	"io"
+	"log/slog"
+	"net/http"
+	"net/http/httptest"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/mathiasbq/hyperguild/ingestion/internal/startupretry"
+)
+
+// fakeIssuer answers OIDC discovery with 502 for the first fail requests, as
+// ingress does while dex is not yet serving on a cold boot (#83), then with a
+// real discovery document and an empty JWKS.
+func fakeIssuer(t *testing.T, fail int32) (*httptest.Server, *atomic.Int32) {
+	t.Helper()
+	var discoveries atomic.Int32
+	mux := http.NewServeMux()
+	var srv *httptest.Server
+	mux.HandleFunc("/.well-known/openid-configuration", func(w http.ResponseWriter, _ *http.Request) {
+		if discoveries.Add(1) <= fail {
+			w.WriteHeader(http.StatusBadGateway)
+			return
+		}
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = io.WriteString(w, `{"issuer":"`+srv.URL+`","jwks_uri":"`+srv.URL+`/keys"}`)
+	})
+	mux.HandleFunc("/keys", func(w http.ResponseWriter, _ *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = io.WriteString(w, `{"keys":[]}`)
+	})
+	srv = httptest.NewServer(mux)
+	t.Cleanup(srv.Close)
+	return srv, &discoveries
+}
+
+var fastPolicy = startupretry.Policy{Attempts: 8, Base: time.Millisecond, Max: 5 * time.Millisecond}
+
+func TestBuildJWTValidatorRetriesDiscovery502sThenSucceeds(t *testing.T) {
+	srv, discoveries := fakeIssuer(t, 3)
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	v, err := buildJWTValidator(ctx, slog.New(slog.DiscardHandler), srv.URL, "brain", fastPolicy)
+	require.NoError(t, err)
+	assert.NotNil(t, v)
+	assert.Equal(t, int32(4), discoveries.Load(), "three 502s, then the successful discovery")
+}
+
+func TestBuildJWTValidatorGivesUpAfterTheBudgetWithTheLastError(t *testing.T) {
+	srv, discoveries := fakeIssuer(t, 1000)
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	_, err := buildJWTValidator(ctx, slog.New(slog.DiscardHandler), srv.URL, "brain", fastPolicy)
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "status 502")
+	assert.Equal(t, int32(8), discoveries.Load(), "bounded: exactly the attempt budget")
+}
+
+func TestBuildJWTValidatorStopsRetryingWhenTheContextIsCancelled(t *testing.T) {
+	srv, _ := fakeIssuer(t, 1000)
+	ctx, cancel := context.WithCancel(context.Background())
+	slow := startupretry.Policy{Attempts: 8, Base: time.Hour, Max: time.Hour}
+
+	done := make(chan error, 1)
+	go func() {
+		_, err := buildJWTValidator(ctx, slog.New(slog.DiscardHandler), srv.URL, "brain", slow)
+		done <- err
+	}()
+	time.Sleep(50 * time.Millisecond)
+	cancel()
+	select {
+	case err := <-done:
+		assert.ErrorIs(t, err, context.Canceled)
+	case <-time.After(3 * time.Second):
+		t.Fatal("did not return after the context was cancelled")
+	}
+}

ingestion/cmd/server/main.go

@@ -381,15 +381,9 @@ func main() {
 	//
 	// The budget stays bounded: retrying forever would turn a misconfiguration
 	// into a process that looks alive and never serves.
-	jwtValidator, err := startupretry.Do(ctx,
+	jwtValidator, err := buildJWTValidator(ctx, logger,
-		startupretry.Policy{Attempts: 8, Base: time.Second, Max: 30 * time.Second},
+		os.Getenv("DEX_ISSUER_URL"), os.Getenv("MCP_AUDIENCE"),
-		func(ctx context.Context) (*chassisauth.JWTValidator, error) {
+		startupretry.Policy{Attempts: 8, Base: time.Second, Max: 30 * time.Second})
-			v, err := chassisauth.NewJWTValidator(ctx, os.Getenv("DEX_ISSUER_URL"), os.Getenv("MCP_AUDIENCE"))
-			if err != nil {
-				logger.Warn("oidc discovery not ready, retrying", "err", err)
-			}
-			return v, err
-		})
 	if err != nil {
 		logger.Error("build jwt validator", "err", err)
 		os.Exit(1)

ingestion/go.mod

@@ -34,9 +34,9 @@ require (
 	github.com/rogpeppe/go-internal v1.15.0 // indirect
 	github.com/spf13/pflag v1.0.5 // indirect
 	github.com/x448/float16 v0.8.4 // indirect
-	golang.org/x/net v0.26.0 // indirect
+	golang.org/x/net v0.55.0 // indirect
 	golang.org/x/oauth2 v0.21.0 // indirect
-	golang.org/x/term v0.28.0 // indirect
+	golang.org/x/term v0.43.0 // indirect
 	golang.org/x/time v0.3.0 // indirect
 	google.golang.org/protobuf v1.34.2 // indirect
 	gopkg.in/evanphx/json-patch.v4 v4.12.0 // indirect
@@ -66,9 +66,9 @@ require (
 	github.com/lestrrat-go/option v1.0.1 // indirect
 	github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2 // indirect
 	github.com/segmentio/asm v1.2.0 // indirect
-	golang.org/x/crypto v0.32.0 // indirect
+	golang.org/x/crypto v0.51.0 // indirect
-	golang.org/x/sync v0.17.0 // indirect
+	golang.org/x/sync v0.21.0 // indirect
-	golang.org/x/sys v0.31.0 // indirect
+	golang.org/x/sys v0.45.0 // indirect
-	golang.org/x/text v0.29.0 // indirect
+	golang.org/x/text v0.39.0 // indirect
 	gopkg.in/yaml.v3 v3.0.1
 )

ingestion/go.sum

@@ -117,42 +117,42 @@ github.com/yuin/goldmark v1.2.1/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9dec
 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
 golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
 golang.org/x/crypto v0.0.0-20200622213623-75b288015ac9/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
-golang.org/x/crypto v0.32.0 h1:euUpcYgM8WcP71gNpTqQCn6rC2t6ULUPiOzfWaXVVfc=
+golang.org/x/crypto v0.51.0 h1:IBPXwPfKxY7cWQZ38ZCIRPI50YLeevDLlLnyC5wRGTI=
-golang.org/x/crypto v0.32.0/go.mod h1:ZnnJkOaASj8g0AjIduWNlq2NRxL0PlBrbKVyZ6V/Ugc=
+golang.org/x/crypto v0.51.0/go.mod h1:8AdwkbraGNABw2kOX6YFPs3WM22XqI4EXEd8g+x7Oc8=
 golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/mod v0.3.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/net v0.0.0-20200226121028-0de0cce0169b/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/net v0.0.0-20201021035429-f5854403a974/go.mod h1:sp8m0HH+o8qH0wwXwYZr8TS3Oi6o0r6Gce1SSxlDquU=
-golang.org/x/net v0.26.0 h1:soB7SVo0PWrY4vPW/+ay0jKDNScG2X9wFeYlXIvJsOQ=
+golang.org/x/net v0.55.0 h1:bcvxaJn3e1U6InsFWt1JUq1aSjnRxLzT2rtD2KfkDF8=
-golang.org/x/net v0.26.0/go.mod h1:5YKkiSynbBIh3p6iOc/vibscux0x38BZDkn8sCUPxHE=
+golang.org/x/net v0.55.0/go.mod h1:L5U2KuzuOe1lY7Z+aWVIKK6qEeJXnXV9yzGA+WCHJww=
 golang.org/x/oauth2 v0.21.0 h1:tsimM75w1tF/uws5rbeHzIWxEqElMehnc+iW793zsZs=
 golang.org/x/oauth2 v0.21.0/go.mod h1:XYTD2NtWslqkgxebSiOHnXEap4TF09sJSc7H1sXbhtI=
 golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20190911185100-cd5d95a43a6e/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sync v0.0.0-20201020160332-67f06af15bc9/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
-golang.org/x/sync v0.17.0 h1:l60nONMj9l5drqw6jlhIELNv9I0A4OFgRsG9k2oT9Ug=
+golang.org/x/sync v0.21.0 h1:HLII4xRRTtCRkxYp4HNFF0Js/Og6q2i++KXbg0gHCwM=
-golang.org/x/sync v0.17.0/go.mod h1:9KTHXmSnoGruLpwFjVSX0lNNA75CykiMECbovNTZqGI=
+golang.org/x/sync v0.21.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200930185726-fdedc70b468f/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.31.0 h1:ioabZlmFYtWhL+TRYpcnNlLwhyxaM9kWTDEmfnprqik=
+golang.org/x/sys v0.45.0 h1:dO4czNzziLiiXplLQgBCEpCvXQ3dnkn0SdaZSYdQ+FY=
-golang.org/x/sys v0.31.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
+golang.org/x/sys v0.45.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
-golang.org/x/term v0.28.0 h1:/Ts8HFuMR2E6IP/jlo7QVLZHggjKQbhu/7H0LJFr3Gg=
+golang.org/x/term v0.43.0 h1:S4RLU2sB31O/NCl+zFN9Aru9A/Cq2aqKpTZJ6B+DwT4=
-golang.org/x/term v0.28.0/go.mod h1:Sw/lC2IAUZ92udQNf3WodGtn4k/XoLyZoh8v/8uiwek=
+golang.org/x/term v0.43.0/go.mod h1:lrhlHNdQJHO+1qVYiHfFKVuVioJIheAc3fBSMFYEIsk=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
-golang.org/x/text v0.29.0 h1:1neNs90w9YzJ9BocxfsQNHKuAT4pkghyXc4nhZ6sJvk=
+golang.org/x/text v0.39.0 h1:UbZz4pLOvn600D6Oh6GGEI6VAmndrEBLv8/6BEXzyus=
-golang.org/x/text v0.29.0/go.mod h1:7MhJOA9CD2qZyOKYazxdYMF85OwPdEr9jTtBpO7ydH4=
+golang.org/x/text v0.39.0/go.mod h1:3UwRclnC2g0TU9x8PZiyfOajCd1zaUNHF9cvqcQZ+ZM=
 golang.org/x/time v0.3.0 h1:rg5rLMjNzMS1RkNLzCG38eapWhnYLFYXDXj2gOlr8j4=
 golang.org/x/time v0.3.0/go.mod h1:tRJNPiyCQ0inRvYxbN9jk5I+vvW/OXSQhTDSoE431IQ=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
 golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
 golang.org/x/tools v0.0.0-20200619180055-7c47624df98f/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
 golang.org/x/tools v0.0.0-20210106214847-113979e3529a/go.mod h1:emZCQorbCU4vsT4fOWvOPXz4eW1wZW4PmDk9uLelYpA=
-golang.org/x/tools v0.36.0 h1:kWS0uv/zsvHEle1LbV5LE8QujrxB3wfQyxHfhOk0Qkg=
+golang.org/x/tools v0.47.0 h1:7Kn5x/d1svx/PzryTsqeoZN4TZwqeH5pGWjefhLi/1Q=
-golang.org/x/tools v0.36.0/go.mod h1:WBDiHKJK8YgLHlcQPYQzNCkUxUypCaa5ZegCVutKm+s=
+golang.org/x/tools v0.47.0/go.mod h1:dFHnyTvFWY212G+h7ZY4Vsp/K3U4/7W9TyVaAul8uCA=
 golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=

ingestion/internal/brain/index.go

@@ -2,12 +2,18 @@ package brain
 
 import (
 	"bufio"
+	"bytes"
+	"errors"
 	"fmt"
+	"io"
+	"io/fs"
 	"os"
 	"path/filepath"
 	"sort"
 	"strings"
 	"time"
+
+	"github.com/mathiasbq/hyperguild/ingestion/internal/brainfs"
 )
 
 // noteEntry is one row in a Wing _index.md.
@@ -27,13 +33,18 @@ func BuildWingIndex(brainDir, wing string) error {
 		return fmt.Errorf("invalid wing %q", wing)
 	}
 	wingDir := filepath.Join(brainDir, "wiki", w)
-	if _, err := os.Stat(wingDir); os.IsNotExist(err) {
+	// Listed and read through the brain root (hyperguild#157): a wing or
+	// hall that is a symlink is not followed, and notes are regular files.
+	bfs, err := brainfs.Open(brainDir)
+	if err != nil {
+		return fmt.Errorf("open brain: %w", err)
+	}
+	defer func() { _ = bfs.Close() }()
+	if _, err := bfs.Lstat("wiki/" + w); errors.Is(err, fs.ErrNotExist) {
 		return nil
-	} else if err != nil {
-		return fmt.Errorf("stat wing: %w", err)
 	}
 
-	entries, err := collectWingEntries(wingDir)
+	entries, err := collectWingEntries(bfs, "wiki/"+w)
 	if err != nil {
 		return err
 	}
@@ -58,51 +69,47 @@ func BuildWingIndex(brainDir, wing string) error {
 
 // BuildAllWingIndexes regenerates _index.md for every wing under brain/wiki/.
 func BuildAllWingIndexes(brainDir string) error {
-	wikiDir := filepath.Join(brainDir, "wiki")
+	bfs, err := brainfs.Open(brainDir)
-	ents, err := os.ReadDir(wikiDir)
+	if err != nil {
-	if os.IsNotExist(err) {
+		return fmt.Errorf("open brain: %w", err)
+	}
+	wings, err := bfs.ReadDirDirs("wiki")
+	_ = bfs.Close()
+	if errors.Is(err, fs.ErrNotExist) {
 		return nil
 	}
 	if err != nil {
 		return fmt.Errorf("read wiki: %w", err)
 	}
-	for _, e := range ents {
+	for _, w := range wings {
-		if !e.IsDir() {
+		if err := BuildWingIndex(brainDir, w); err != nil {
-			continue
+			return fmt.Errorf("index %s: %w", w, err)
-		}
-		if err := BuildWingIndex(brainDir, e.Name()); err != nil {
-			return fmt.Errorf("index %s: %w", e.Name(), err)
 		}
 	}
 	return nil
 }
 
-func collectWingEntries(wingDir string) ([]noteEntry, error) {
+func collectWingEntries(bfs *brainfs.Brain, wingRel string) ([]noteEntry, error) {
 	var out []noteEntry
-	ents, err := os.ReadDir(wingDir)
+	halls, err := bfs.ReadDirDirs(wingRel)
 	if err != nil {
 		return nil, fmt.Errorf("read wing: %w", err)
 	}
-	for _, hallEnt := range ents {
+	for _, hall := range halls {
-		if !hallEnt.IsDir() {
-			continue
-		}
-		hall := hallEnt.Name()
 		if !IsValidHall(hall) {
 			continue
 		}
-		hallDir := filepath.Join(wingDir, hall)
+		hallRel := wingRel + "/" + hall
-		notes, err := os.ReadDir(hallDir)
+		names, err := bfs.ReadDirRegular(hallRel)
 		if err != nil {
 			return nil, fmt.Errorf("read hall %s: %w", hall, err)
 		}
-		for _, n := range notes {
+		for _, name := range names {
-			if n.IsDir() || !strings.HasSuffix(n.Name(), ".md") || n.Name() == "_index.md" {
+			if !strings.HasSuffix(name, ".md") || name == "_index.md" {
 				continue
 			}
-			slug := strings.TrimSuffix(n.Name(), ".md")
+			slug := strings.TrimSuffix(name, ".md")
-			full := filepath.Join(hallDir, n.Name())
+			title, created := readTitleAndCreatedIn(bfs, hallRel+"/"+name, slug)
-			title, created := readTitleAndCreated(full, slug)
 			out = append(out, noteEntry{Hall: hall, Slug: slug, Title: title, Created: created})
 		}
 	}
@@ -117,9 +124,40 @@ func readTitleAndCreated(path, slug string) (string, string) {
 		return slug, ""
 	}
 	defer func() { _ = f.Close() }()
+	title, created := parseTitleAndCreated(f)
+	if created == "" {
+		if info, err := os.Stat(path); err == nil {
+			created = info.ModTime().UTC().Format("2006-01-02")
+		}
+	}
+	return titleOrSlug(title, slug), created
+}
 
-	title, created := "", ""
+// readTitleAndCreatedIn is readTitleAndCreated through the brain root,
-	scanner := bufio.NewScanner(f)
+// regular files only (hyperguild#157).
+func readTitleAndCreatedIn(bfs *brainfs.Brain, rel, slug string) (string, string) {
+	content, err := bfs.ReadRegular(rel)
+	if err != nil {
+		return slug, ""
+	}
+	title, created := parseTitleAndCreated(bytes.NewReader(content))
+	if created == "" {
+		if info, err := bfs.Lstat(rel); err == nil {
+			created = info.ModTime().UTC().Format("2006-01-02")
+		}
+	}
+	return titleOrSlug(title, slug), created
+}
+
+func titleOrSlug(title, slug string) string {
+	if title == "" {
+		return strings.ReplaceAll(slug, "-", " ")
+	}
+	return title
+}
+
+func parseTitleAndCreated(r io.Reader) (title, created string) {
+	scanner := bufio.NewScanner(r)
 	inFrontmatter := false
 	for scanner.Scan() {
 		line := scanner.Text()
@@ -149,13 +187,5 @@ func readTitleAndCreated(path, slug string) (string, string) {
 			}
 		}
 	}
-	if title == "" {
-		title = strings.ReplaceAll(slug, "-", " ")
-	}
-	if created == "" {
-		if info, err := os.Stat(path); err == nil {
-			created = info.ModTime().UTC().Format("2006-01-02")
-		}
-	}
 	return title, created
 }

ingestion/internal/brain/index_root_test.go

@@ -0,0 +1,132 @@
+package brain_test
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/mathiasbq/hyperguild/ingestion/internal/brain"
+	"github.com/mathiasbq/hyperguild/ingestion/internal/brainfs/brainfstest"
+)
+
+func readIndex(t *testing.T, brainDir, wing string) string {
+	t.Helper()
+	b, _ := os.ReadFile(filepath.Join(brainDir, "wiki", wing, "_index.md"))
+	return string(b)
+}
+
+// hyperguild#157 slice 5: BuildWingIndex listed wiki/<wing>/<hall>/ with
+// os.ReadDir and opened each note by path, so a symlinked note put the
+// target's title into _index.md, and a FIFO hung the index build.
+func TestBuildWingIndexRefusesSymlinkedNotesAndSurvivesSwapRaces(t *testing.T) {
+	b := t.TempDir()
+	facts := filepath.Join(b, "wiki", "w", "facts")
+	require.NoError(t, os.MkdirAll(facts, 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(facts, "ok.md"), []byte("---\ntitle: zqinside\n---\n"), 0o644))
+	require.NoError(t, os.Symlink(brainfstest.Outside(t), filepath.Join(facts, "leak.md")))
+	inBrain := brainfstest.WriteSentinelAt(t, filepath.Join(b, "raw", "s.md"))
+	require.NoError(t, os.Symlink(filepath.Join("..", "..", "..", "raw", "s.md"), filepath.Join(facts, "inlink.md")))
+
+	require.NoError(t, brain.BuildWingIndex(b, "w"))
+	idx := readIndex(t, b, "w")
+	assert.Contains(t, idx, "zqinside", "positive control")
+	assert.NotContains(t, idx, brainfstest.Sentinel)
+	assert.NotContains(t, idx, "leak")
+	assert.NotContains(t, idx, "inlink")
+
+	p := filepath.Join(facts, "r.md")
+	read := func() string {
+		_ = brain.BuildWingIndex(b, "w")
+		return readIndex(t, b, "w")
+	}
+	brainfstest.AssertNoLeak(t, p, "---\ntitle: zqinside\n---\n", 1500*time.Millisecond, read)
+	brainfstest.AssertNoLeakFrom(t, p, "---\ntitle: zqinside\n---\n", inBrain, 1500*time.Millisecond, read)
+	brainfstest.AssertNoHang(t, p, "---\ntitle: zqinside\n---\n", 1500*time.Millisecond, 6*time.Second, func() { _ = read() })
+}
+
+// A symlinked hall is skipped while a real hall next to it is indexed; a
+// symlinked wing is refused and nothing is written into its target.
+func TestBuildWingIndexSkipsASymlinkedHallAndRefusesASymlinkedWing(t *testing.T) {
+	b := t.TempDir()
+	require.NoError(t, os.MkdirAll(filepath.Join(b, "raw", "facts"), 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(b, "raw", "facts", "r.md"), []byte("---\ntitle: zqrawonly\n---\n"), 0o644))
+	require.NoError(t, os.MkdirAll(filepath.Join(b, "wiki", "w", "decisions"), 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(b, "wiki", "w", "decisions", "d.md"), []byte("---\ntitle: zqrealhall\n---\n"), 0o644))
+	require.NoError(t, os.Symlink(filepath.Join("..", "..", "raw", "facts"), filepath.Join(b, "wiki", "w", "facts")))
+	require.NoError(t, os.Symlink(filepath.Join("..", "raw"), filepath.Join(b, "wiki", "alias")))
+
+	require.NoError(t, brain.BuildWingIndex(b, "w"))
+	idx, err := os.ReadFile(filepath.Join(b, "wiki", "w", "_index.md"))
+	require.NoError(t, err, "the wing index is written")
+	assert.Contains(t, string(idx), "zqrealhall", "the real hall is indexed")
+	assert.NotContains(t, string(idx), "zqrawonly", "the symlinked hall is not")
+
+	assert.Error(t, brain.BuildWingIndex(b, "alias"))
+	_, err = os.Stat(filepath.Join(b, "raw", "_index.md"))
+	assert.True(t, os.IsNotExist(err), "an index was written into the symlinked wing's target")
+}
+
+// BuildAllWingIndexes indexes the real wings and skips a symlinked one.
+func TestBuildAllWingIndexesSkipsASymlinkedWing(t *testing.T) {
+	b := t.TempDir()
+	require.NoError(t, os.MkdirAll(filepath.Join(b, "wiki", "w", "facts"), 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(b, "wiki", "w", "facts", "n.md"), []byte("---\ntitle: zqrealwing\n---\n"), 0o644))
+	require.NoError(t, os.MkdirAll(filepath.Join(b, "raw", "facts"), 0o755))
+	require.NoError(t, os.Symlink(filepath.Join("..", "raw"), filepath.Join(b, "wiki", "alias")))
+
+	require.NoError(t, brain.BuildAllWingIndexes(b))
+	idx, err := os.ReadFile(filepath.Join(b, "wiki", "w", "_index.md"))
+	require.NoError(t, err)
+	assert.Contains(t, string(idx), "zqrealwing")
+	_, err = os.Stat(filepath.Join(b, "raw", "_index.md"))
+	assert.True(t, os.IsNotExist(err))
+}
+
+// A link to a symlinked note does not resolve: only regular files count as
+// entries, so StripUnresolvedLinks unwraps it.
+func TestStripUnresolvedLinksDoesNotCountASymlinkedNoteAsAnEntry(t *testing.T) {
+	b := t.TempDir()
+	facts := filepath.Join(b, "wiki", "w", "facts")
+	require.NoError(t, os.MkdirAll(facts, 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(facts, "real.md"), []byte("x"), 0o644))
+	brainfstest.WriteSentinelAt(t, filepath.Join(b, "raw", "s.md"))
+	require.NoError(t, os.Symlink(filepath.Join("..", "..", "..", "raw", "s.md"), filepath.Join(facts, "ghost.md")))
+	out, unlinked, err := brain.StripUnresolvedLinks(b, "see [[real]] and [[ghost]]")
+	require.NoError(t, err)
+	assert.Contains(t, out, "[[real]]", "positive control")
+	assert.NotContains(t, out, "[[ghost]]")
+	assert.Contains(t, unlinked, "ghost")
+}
+
+// A linked tree (knowledge/ or wiki/) that is not a real directory is
+// skipped, not an error: StripUnresolvedLinks runs before every brain_write,
+// and an error there loses the learning.
+func TestStripUnresolvedLinksSkipsALinkedTreeThatIsNotARealDirectory(t *testing.T) {
+	for name, mk := range map[string]func(t *testing.T, b string){
+		"symlink to an in-brain dir": func(t *testing.T, b string) {
+			require.NoError(t, os.MkdirAll(filepath.Join(b, "raw", "k"), 0o755))
+			require.NoError(t, os.Symlink(filepath.Join("raw", "k"), filepath.Join(b, "knowledge")))
+		},
+		"dangling symlink": func(t *testing.T, b string) {
+			require.NoError(t, os.Symlink("nowhere", filepath.Join(b, "knowledge")))
+		},
+		"regular file": func(t *testing.T, b string) {
+			require.NoError(t, os.WriteFile(filepath.Join(b, "knowledge"), []byte("x"), 0o644))
+		},
+	} {
+		t.Run(name, func(t *testing.T) {
+			b := t.TempDir()
+			require.NoError(t, os.MkdirAll(filepath.Join(b, "wiki", "w", "facts"), 0o755))
+			require.NoError(t, os.WriteFile(filepath.Join(b, "wiki", "w", "facts", "real.md"), []byte("x"), 0o644))
+			mk(t, b)
+			out, _, err := brain.StripUnresolvedLinks(b, "learning [[real]] and [[gone]]")
+			require.NoError(t, err)
+			assert.Contains(t, out, "[[real]]", "links into wiki/ still resolve")
+			assert.NotContains(t, out, "[[gone]]")
+		})
+	}
+}

ingestion/internal/brain/linkcheck.go

@@ -1,12 +1,16 @@
 package brain
 
 import (
+	"errors"
 	"fmt"
-	"os"
+	"io/fs"
-	"path/filepath"
+	"log/slog"
+	"path"
 	"regexp"
 	"sort"
 	"strings"
+
+	"github.com/mathiasbq/hyperguild/ingestion/internal/brainfs"
 )
 
 // wikilinkRE matches [[target]] and [[target|Display Text]].
@@ -98,29 +102,40 @@ var linkedTrees = []string{"knowledge", "wiki"}
 // one the audit will also resolve.
 func entryIndex(brainDir string) (map[string]struct{}, error) {
 	index := map[string]struct{}{}
-	for _, dir := range linkedTrees {
+	// Walked through the brain root (hyperguild#157): only regular files
-		root := filepath.Join(brainDir, dir)
+	// count as entries, so a symlinked .md name cannot make a link resolve.
-		if _, err := os.Stat(root); os.IsNotExist(err) {
+	bfs, err := brainfs.Open(brainDir)
-			continue
-		} else if err != nil {
-			return nil, fmt.Errorf("stat %s: %w", dir, err)
-		}
-		err := filepath.WalkDir(root, func(path string, d os.DirEntry, err error) error {
 	if err != nil {
-				return err
+		return nil, fmt.Errorf("open brain: %w", err)
 	}
-			if d.IsDir() || !strings.HasSuffix(path, ".md") {
+	defer func() { _ = bfs.Close() }()
+	for _, dir := range linkedTrees {
+		if _, err := bfs.Lstat(dir); errors.Is(err, fs.ErrNotExist) {
+			continue
+		}
+		var walkErr error
+		err := bfs.WalkRegular(dir, func(rel string) error {
+			if !strings.HasSuffix(rel, ".md") {
 				return nil
 			}
-			rel, err := filepath.Rel(brainDir, path)
+			index[strings.TrimSuffix(path.Base(rel), ".md")] = struct{}{}
-			if err != nil {
-				return err
-			}
-			rel = filepath.ToSlash(rel)
-			index[strings.TrimSuffix(filepath.Base(rel), ".md")] = struct{}{}
 			index[strings.TrimPrefix(strings.TrimSuffix(rel, ".md"), "wiki/")] = struct{}{}
 			return nil
+		}, func(rel string, err error) {
+			// A linked tree that is not a real directory (a symlink, a
+			// file) is skipped, as filepath.WalkDir did: this runs before
+			// every brain_write, and an error here loses the learning.
+			if rel == dir && errors.Is(err, brainfs.ErrNotRegular) {
+				slog.Warn("link check: skipping a linked tree that is not a real directory", "dir", dir)
+				return
+			}
+			if walkErr == nil {
+				walkErr = fmt.Errorf("%s: %w", rel, err)
+			}
 		})
+		if err == nil {
+			err = walkErr
+		}
 		if err != nil {
 			return nil, fmt.Errorf("walk %s: %w", dir, err)
 		}

ingestion/internal/brainfs/brainfs.go

@@ -82,6 +82,49 @@ func (b *Brain) realDir(rel string) error {
 	return nil
 }
 
+// ReadDirRegular lists the regular files directly in a brain-relative
+// directory, by name. Every component of relDir must be a real directory
+// (ErrNotRegular otherwise); symlinks, FIFOs, devices and subdirectories in
+// it are left out.
+func (b *Brain) ReadDirRegular(relDir string) ([]string, error) {
+	relDir = filepath.ToSlash(relDir)
+	if err := b.realDir(relDir); err != nil {
+		return nil, err
+	}
+	ents, err := fs.ReadDir(b.root.FS(), relDir)
+	if err != nil {
+		return nil, err
+	}
+	var names []string
+	for _, e := range ents {
+		if e.Type().IsRegular() {
+			names = append(names, e.Name())
+		}
+	}
+	return names, nil
+}
+
+// ReadDirDirs lists the real subdirectories directly in a brain-relative
+// directory, by name. Every component of relDir must be a real directory
+// (ErrNotRegular otherwise); symlinks to directories are left out.
+func (b *Brain) ReadDirDirs(relDir string) ([]string, error) {
+	relDir = filepath.ToSlash(relDir)
+	if err := b.realDir(relDir); err != nil {
+		return nil, err
+	}
+	ents, err := fs.ReadDir(b.root.FS(), relDir)
+	if err != nil {
+		return nil, err
+	}
+	var names []string
+	for _, e := range ents {
+		if e.Type().IsDir() {
+			names = append(names, e.Name())
+		}
+	}
+	return names, nil
+}
+
 // ReadRegular reads a brain-relative path, refusing anything that is not a
 // regular file, both before the open and on the opened handle.
 func (b *Brain) ReadRegular(rel string) ([]byte, error) {
@@ -92,7 +135,7 @@ func (b *Brain) ReadRegular(rel string) ([]byte, error) {
 	if !fi.Mode().IsRegular() {
 		return nil, fmt.Errorf("%s: %w", rel, ErrNotRegular)
 	}
-	return b.readOpened(rel)
+	return b.readOpenedMax(rel, -1, fi)
 }
 
 // readOpened opens rel non-blocking and reads it only if the OPENED handle is
@@ -106,7 +149,7 @@ func (b *Brain) ReadRegular(rel string) ([]byte, error) {
 // cannot create one; fs.protected_hardlinks=1 and the pod's secrets are on a
 // different filesystem from the brain volume, so a local caller cannot link
 // them in. Refusing Nlink > 1 is the next step if either changes.
-func (b *Brain) readOpened(rel string) ([]byte, error) { return b.readOpenedMax(rel, -1) }
+func (b *Brain) readOpened(rel string) ([]byte, error) { return b.readOpenedMax(rel, -1, nil) }
 
 // ReadRegularMax is ReadRegular with a size limit, checked on the opened
 // handle before reading (max < 0 means no limit).
@@ -118,10 +161,14 @@ func (b *Brain) ReadRegularMax(rel string, max int64) ([]byte, error) {
 	if !fi.Mode().IsRegular() {
 		return nil, fmt.Errorf("%s: %w", rel, ErrNotRegular)
 	}
-	return b.readOpenedMax(rel, max)
+	return b.readOpenedMax(rel, max, fi)
 }
 
-func (b *Brain) readOpenedMax(rel string, max int64) ([]byte, error) {
+// readOpenedMax opens rel and reads it if the opened handle is a regular file
+// and, when lfi is set, the same file the caller Lstat-ed: os.Root follows an
+// in-root final symlink even with O_NOFOLLOW, so a symlink to another brain
+// file swapped in between the Lstat and the open is refused only here.
+func (b *Brain) readOpenedMax(rel string, max int64, lfi fs.FileInfo) ([]byte, error) {
 	f, err := b.root.OpenFile(rel, os.O_RDONLY|syscall.O_NONBLOCK, 0)
 	if err != nil {
 		return nil, err
@@ -134,6 +181,9 @@ func (b *Brain) readOpenedMax(rel string, max int64) ([]byte, error) {
 	if !hfi.Mode().IsRegular() {
 		return nil, fmt.Errorf("%s: %w", rel, ErrNotRegular)
 	}
+	if lfi != nil && !os.SameFile(lfi, hfi) {
+		return nil, fmt.Errorf("%s: changed while opening: %w", rel, ErrNotRegular)
+	}
 	if max >= 0 && hfi.Size() > max {
 		return nil, fmt.Errorf("%s is %d bytes (limit %d): %w", rel, hfi.Size(), max, ErrTooLarge)
 	}

ingestion/internal/brainfs/brainfs_test.go

@@ -6,6 +6,8 @@ import (
 	"io/fs"
 	"os"
 	"path/filepath"
+	"sync"
+	"sync/atomic"
 	"syscall"
 	"testing"
 	"time"
@@ -189,3 +191,92 @@ func TestWalkRegularReportsASymlinkedIntermediateComponentOfTheWalkRoot(t *testi
 	require.Len(t, errs, 1)
 	assert.ErrorIs(t, errs[0], ErrNotRegular)
 }
+
+func TestReadDirRegularListsOnlyRegularFilesAndRefusesSymlinkedComponents(t *testing.T) {
+	dir := t.TempDir()
+	require.NoError(t, os.MkdirAll(filepath.Join(dir, "a", "b", "sub"), 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(dir, "a", "b", "n.md"), []byte("n"), 0o644))
+	require.NoError(t, os.WriteFile(filepath.Join(dir, "other.md"), []byte("o"), 0o644))
+	require.NoError(t, os.Symlink(filepath.Join("..", "..", "other.md"), filepath.Join(dir, "a", "b", "l.md")))
+	require.NoError(t, syscall.Mkfifo(filepath.Join(dir, "a", "b", "f.md"), 0o644))
+	require.NoError(t, os.Symlink("a", filepath.Join(dir, "alias")))
+	require.NoError(t, os.MkdirAll(filepath.Join(dir, "x"), 0o755))
+	require.NoError(t, os.Symlink(filepath.Join("..", "a"), filepath.Join(dir, "x", "mid")))
+	b, err := Open(dir)
+	require.NoError(t, err)
+	defer func() { _ = b.Close() }()
+
+	names, err := b.ReadDirRegular("a/b")
+	require.NoError(t, err)
+	assert.Equal(t, []string{"n.md"}, names)
+	for _, rel := range []string{"alias/b", "x/mid/b"} {
+		_, err := b.ReadDirRegular(rel)
+		assert.ErrorIs(t, err, ErrNotRegular, rel)
+	}
+}
+
+// A final component swapped for a symlink to another regular file in the
+// brain between the Lstat and the open: os.Root follows in-root links, so
+// only comparing the two (os.SameFile) refuses it.
+func TestReadRegularNeverReturnsAnInBrainFileSwappedInBehindASymlink(t *testing.T) {
+	dir := t.TempDir()
+	require.NoError(t, os.WriteFile(filepath.Join(dir, "s.md"), []byte("zqsamefilesentinel"), 0o644))
+	p := filepath.Join(dir, "r.md")
+	require.NoError(t, os.WriteFile(p, []byte("inside"), 0o644))
+	b, err := Open(dir)
+	require.NoError(t, err)
+	defer func() { _ = b.Close() }()
+
+	var stop atomic.Bool
+	var wg sync.WaitGroup
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		tmp := p + ".swap"
+		for i := 0; !stop.Load(); i++ {
+			_ = os.Remove(tmp)
+			if i%2 == 0 {
+				_ = os.WriteFile(tmp, []byte("inside"), 0o644)
+			} else {
+				_ = os.Symlink("s.md", tmp)
+			}
+			_ = os.Rename(tmp, p)
+		}
+	}()
+	leaks, ok := 0, 0
+	for deadline := time.Now().Add(1500 * time.Millisecond); time.Now().Before(deadline); {
+		got, err := b.ReadRegular("r.md")
+		if err == nil {
+			ok++
+		}
+		if string(got) == "zqsamefilesentinel" {
+			leaks++
+		}
+		got, _ = b.ReadRegularMax("r.md", 1<<20)
+		if string(got) == "zqsamefilesentinel" {
+			leaks++
+		}
+	}
+	stop.Store(true)
+	wg.Wait()
+	assert.Zero(t, leaks)
+	assert.Positive(t, ok, "positive control: the regular file is read")
+}
+
+func TestReadDirDirsListsOnlyRealSubdirsAndRefusesSymlinkedComponents(t *testing.T) {
+	dir := t.TempDir()
+	require.NoError(t, os.MkdirAll(filepath.Join(dir, "a", "real"), 0o755))
+	require.NoError(t, os.MkdirAll(filepath.Join(dir, "other"), 0o755))
+	require.NoError(t, os.Symlink(filepath.Join("..", "other"), filepath.Join(dir, "a", "linked")))
+	require.NoError(t, os.WriteFile(filepath.Join(dir, "a", "f.md"), []byte("f"), 0o644))
+	require.NoError(t, os.Symlink("a", filepath.Join(dir, "alias")))
+	b, err := Open(dir)
+	require.NoError(t, err)
+	defer func() { _ = b.Close() }()
+
+	names, err := b.ReadDirDirs("a")
+	require.NoError(t, err)
+	assert.Equal(t, []string{"real"}, names)
+	_, err = b.ReadDirDirs("alias")
+	assert.ErrorIs(t, err, ErrNotRegular)
+}

ingestion/internal/brainfs/brainfstest/brainfstest.go

@@ -7,6 +7,7 @@
 package brainfstest
 
 import (
+	"fmt"
 	"os"
 	"path/filepath"
 	"strings"
@@ -20,16 +21,23 @@ import (
 // Sentinel marks the outside file's contents.
 const Sentinel = "zqbrainfssentinelonefiftyseven"
 
+// outsideContent carries Sentinel in every field a reader might extract: the
+// frontmatter title and aliases, a heading, a wikilink and the body.
+const outsideContent = "---\ntitle: " + Sentinel + "\naliases:\n  - " + Sentinel + "\n---\n# " + Sentinel +
+	"\n\n" + Sentinel + " [[" + Sentinel + "]]\n"
+
 // Outside writes a file outside any brain dir containing Sentinel.
 func Outside(t *testing.T) string {
 	t.Helper()
 	p := filepath.Join(t.TempDir(), "secret.md")
-	if err := os.WriteFile(p, []byte("# secret\n\n"+Sentinel), 0o644); err != nil {
+	if err := os.WriteFile(p, []byte(outsideContent), 0o644); err != nil {
 		t.Fatal(err)
 	}
 	return p
 }
 
+var swapSeq atomic.Int64
+
 // StartSwap renames, in a loop, a regular note (content inside) and either
 // a symlink to outside or a FIFO into path, until the test ends. FIFOs get a
 // writer held open until the swapper exits, so a reader that skips the
@@ -42,7 +50,15 @@ func StartSwap(t *testing.T, path, inside, outside string, fifo bool) {
 	wg.Add(1)
 	go func() {
 		defer wg.Done()
-		tmp := path + ".swap"
+		// A name of its own: swappers on one path share a test, and one
+		// opening another's FIFO for write would block forever.
+		tmp := fmt.Sprintf("%s.swap%d", path, swapSeq.Add(1))
+		// A relative link: os.Root refuses every absolute link on its own, so
+		// an absolute in-brain target would test nothing.
+		link := outside
+		if rel, err := filepath.Rel(filepath.Dir(path), outside); err == nil {
+			link = rel
+		}
 		var writers []*os.File
 		defer func() {
 			for _, w := range writers {
@@ -61,7 +77,7 @@ func StartSwap(t *testing.T, path, inside, outside string, fifo bool) {
 					}
 				}
 			default:
-				_ = os.Symlink(outside, tmp)
+				_ = os.Symlink(link, tmp)
 			}
 			_ = os.Rename(tmp, path)
 		}
@@ -69,11 +85,33 @@ func StartSwap(t *testing.T, path, inside, outside string, fifo bool) {
 	t.Cleanup(func() { stop.Store(true); wg.Wait() })
 }
 
+// WriteSentinelAt writes a file carrying Sentinel at p (creating its
+// directory), for an in-brain symlink target.
+func WriteSentinelAt(t *testing.T, p string) string {
+	t.Helper()
+	if err := os.MkdirAll(filepath.Dir(p), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(p, []byte(outsideContent), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	return p
+}
+
 // AssertNoLeak loops read for d while a symlink swapper runs on path, and
-// fails if any output contains Sentinel.
+// fails if any output contains Sentinel. The symlink points outside the
+// brain; os.Root refuses that on its own, so pair it with AssertNoLeakFrom
+// and an in-brain target.
 func AssertNoLeak(t *testing.T, path, inside string, d time.Duration, read func() string) {
 	t.Helper()
-	StartSwap(t, path, inside, Outside(t), false)
+	AssertNoLeakFrom(t, path, inside, Outside(t), d, read)
+}
+
+// AssertNoLeakFrom is AssertNoLeak with the symlink pointing at target, e.g.
+// a file elsewhere in the brain written with WriteSentinelAt.
+func AssertNoLeakFrom(t *testing.T, path, inside, target string, d time.Duration, read func() string) {
+	t.Helper()
+	StartSwap(t, path, inside, target, false)
 	leaks, iters := 0, 0
 	for deadline := time.Now().Add(d); time.Now().Before(deadline); iters++ {
 		leaks += strings.Count(read(), Sentinel)

ingestion/internal/graphsync/graphsync.go

@@ -12,10 +12,14 @@ package graphsync
 
 import (
 	"context"
+	"errors"
 	"fmt"
-	"os"
+	"io/fs"
+	"path"
 	"path/filepath"
+	"strings"
 
+	"github.com/mathiasbq/hyperguild/ingestion/internal/brainfs"
 	"github.com/mathiasbq/hyperguild/ingestion/internal/graph"
 	"github.com/mathiasbq/hyperguild/ingestion/internal/graphstore"
 )
@@ -46,8 +50,25 @@ func IndexDoc(ctx context.Context, store Store, brainDir, relPath string) error
 	if relPath == "" {
 		return nil
 	}
-	abs := filepath.Join(brainDir, filepath.FromSlash(relPath))
+	// Callers pass caller-supplied paths (brain_tunnel, brain_update): index
-	content, err := os.ReadFile(abs)
+	// only a clean path under wiki/ or knowledge/.
+	if path.Clean(relPath) != relPath || !filepath.IsLocal(relPath) ||
+		(!strings.HasPrefix(relPath, "wiki/") && !strings.HasPrefix(relPath, "knowledge/")) {
+		return fmt.Errorf("index %q: not a clean wiki/ or knowledge/ path", relPath)
+	}
+	bfs, err := brainfs.Open(brainDir)
+	if err != nil {
+		return fmt.Errorf("open brain: %w", err)
+	}
+	defer func() { _ = bfs.Close() }()
+	return indexDoc(ctx, store, bfs, relPath)
+}
+
+// indexDoc reads relPath through the brain root, regular files only
+// (hyperguild#157): the brain is a Syncthing folder, and os.ReadFile
+// followed a synced symlink into the graph and blocked on a FIFO.
+func indexDoc(ctx context.Context, store Store, bfs *brainfs.Brain, relPath string) error {
+	content, err := bfs.ReadRegular(relPath)
 	if err != nil {
 		return fmt.Errorf("read %q: %w", relPath, err)
 	}
@@ -77,33 +98,34 @@ func BackfillFromBrainDir(ctx context.Context, store Store, brainDir string) (in
 	if store == nil {
 		return 0, nil
 	}
-	roots := []string{"wiki", "knowledge"}
+	bfs, err := brainfs.Open(brainDir)
-	for _, root := range roots {
+	if err != nil {
-		base := filepath.Join(brainDir, root)
+		return 0, fmt.Errorf("open brain: %w", err)
-		if _, err := os.Stat(base); os.IsNotExist(err) {
+	}
+	defer func() { _ = bfs.Close() }()
+	for _, root := range []string{"wiki", "knowledge"} {
+		if _, err := bfs.Lstat(root); errors.Is(err, fs.ErrNotExist) {
 			continue
 		}
-		err := filepath.WalkDir(base, func(path string, d os.DirEntry, walkErr error) error {
+		// Any walk error, including a root that is a symlink, aborts.
-			if walkErr != nil {
+		var walkErr error
-				return walkErr
+		err := bfs.WalkRegular(root, func(rel string) error {
-			}
+			if path.Ext(rel) != ".md" {
-			if d.IsDir() {
 				return nil
 			}
-			if filepath.Ext(path) != ".md" {
+			if err := indexDoc(ctx, store, bfs, rel); err != nil {
-				return nil
-			}
-			rel, relErr := filepath.Rel(brainDir, path)
-			if relErr != nil {
-				return fmt.Errorf("rel %q: %w", path, relErr)
-			}
-			rel = filepath.ToSlash(rel)
-			if err := IndexDoc(ctx, store, brainDir, rel); err != nil {
 				return fmt.Errorf("index %q: %w", rel, err)
 			}
 			indexed++
 			return nil
+		}, func(rel string, err error) {
+			if walkErr == nil {
+				walkErr = fmt.Errorf("%s: %w", rel, err)
+			}
 		})
+		if err == nil {
+			err = walkErr
+		}
 		if err != nil {
 			return indexed, fmt.Errorf("walk %s: %w", root, err)
 		}

ingestion/internal/graphsync/graphsync_root_test.go

@@ -0,0 +1,92 @@
+package graphsync
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/mathiasbq/hyperguild/ingestion/internal/brainfs/brainfstest"
+)
+
+func (f *fakeStore) dump() string {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return fmt.Sprintf("%+v %+v", f.upserts, f.replaces)
+}
+
+// hyperguild#157 slice 4: IndexDoc and BackfillFromBrainDir read notes with
+// os.ReadFile, so a symlinked note put the target's title and wikilinks into
+// the graph, and a FIFO hung the read.
+func TestIndexDocRefusesSymlinksAndSurvivesSwapRaces(t *testing.T) {
+	brain := t.TempDir()
+	facts := filepath.Join(brain, "wiki", "w", "facts")
+	require.NoError(t, os.MkdirAll(facts, 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(facts, "ok.md"), []byte("---\ntitle: zqinside\n---\nbody [[other]]\n"), 0o644))
+	require.NoError(t, os.Symlink(brainfstest.Outside(t), filepath.Join(facts, "leak.md")))
+
+	store := newFakeStore()
+	require.NoError(t, IndexDoc(context.Background(), store, brain, "wiki/w/facts/ok.md"))
+	assert.Contains(t, store.dump(), "zqinside", "positive control: a regular note is indexed")
+	assert.Error(t, IndexDoc(context.Background(), store, brain, "wiki/w/facts/leak.md"))
+	// A link to a file elsewhere in the brain: os.Root allows it, only the
+	// regular-file check refuses it.
+	inBrain := brainfstest.WriteSentinelAt(t, filepath.Join(brain, "raw", "s.md"))
+	require.NoError(t, os.Symlink(filepath.Join("..", "..", "..", "raw", "s.md"), filepath.Join(facts, "inlink.md")))
+	assert.Error(t, IndexDoc(context.Background(), store, brain, "wiki/w/facts/inlink.md"))
+	assert.NotContains(t, store.dump(), brainfstest.Sentinel)
+
+	p := filepath.Join(facts, "r.md")
+	read := func() string {
+		s := newFakeStore()
+		_ = IndexDoc(context.Background(), s, brain, "wiki/w/facts/r.md")
+		return s.dump()
+	}
+	brainfstest.AssertNoLeak(t, p, "---\ntitle: zqinside\n---\n", 1500*time.Millisecond, read)
+	brainfstest.AssertNoLeakFrom(t, p, "---\ntitle: zqinside\n---\n", inBrain, 1500*time.Millisecond, read)
+	brainfstest.AssertNoHang(t, p, "---\ntitle: zqinside\n---\n", 1500*time.Millisecond, 6*time.Second, func() { _ = read() })
+}
+
+func TestBackfillSkipsSymlinkedNotesAndAbortsOnASymlinkedRoot(t *testing.T) {
+	brain := t.TempDir()
+	facts := filepath.Join(brain, "wiki", "w", "facts")
+	require.NoError(t, os.MkdirAll(facts, 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(facts, "ok.md"), []byte("---\ntitle: zqinside\n---\n"), 0o644))
+	require.NoError(t, os.Symlink(brainfstest.Outside(t), filepath.Join(facts, "leak.md")))
+	require.NoError(t, os.MkdirAll(filepath.Join(brain, "raw"), 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(brain, "raw", "r.md"), []byte("---\ntitle: zqrawonly\n---\n"), 0o644))
+	require.NoError(t, os.Symlink(filepath.Join("..", "..", "..", "raw", "r.md"), filepath.Join(facts, "inlink.md")))
+	require.NoError(t, os.Symlink(".", filepath.Join(brain, "knowledge")))
+
+	store := newFakeStore()
+	_, err := BackfillFromBrainDir(context.Background(), store, brain)
+	assert.Error(t, err, "a symlinked root aborts the backfill")
+	assert.NotContains(t, store.dump(), brainfstest.Sentinel)
+	assert.NotContains(t, store.dump(), "zqrawonly")
+
+	require.NoError(t, os.Remove(filepath.Join(brain, "knowledge")))
+	store = newFakeStore()
+	n, err := BackfillFromBrainDir(context.Background(), store, brain)
+	require.NoError(t, err)
+	assert.Equal(t, 1, n, "only the regular note is indexed")
+	assert.Contains(t, store.dump(), "zqinside")
+	assert.NotContains(t, store.dump(), brainfstest.Sentinel)
+}
+
+// IndexDoc gets caller-supplied paths (brain_tunnel, brain_update): only a
+// clean path under wiki/ or knowledge/ is indexed.
+func TestIndexDocRefusesPathsOutsideWikiAndKnowledgeAndUncleanPaths(t *testing.T) {
+	brain := t.TempDir()
+	brainfstest.WriteSentinelAt(t, filepath.Join(brain, "raw", "secret.md"))
+	require.NoError(t, os.MkdirAll(filepath.Join(brain, "wiki", "w"), 0o755))
+	for _, p := range []string{"raw/secret.md", "wiki/w/../../raw/secret.md", "./raw/secret.md", "/raw/secret.md"} {
+		store := newFakeStore()
+		assert.Error(t, IndexDoc(context.Background(), store, brain, p), p)
+		assert.NotContains(t, store.dump(), brainfstest.Sentinel, p)
+	}
+}

ingestion/internal/wiki/index.go

@@ -7,6 +7,7 @@ import (
 	"path/filepath"
 	"strings"
 
+	"github.com/mathiasbq/hyperguild/ingestion/internal/brainfs"
 	"github.com/mathiasbq/hyperguild/ingestion/internal/textutil"
 )
 
@@ -16,6 +17,11 @@ func RebuildIndex(brainDir, date string) error {
 	if err != nil {
 		return fmt.Errorf("load inventory: %w", err)
 	}
+	bfs, err := brainfs.Open(brainDir)
+	if err != nil {
+		return fmt.Errorf("open brain: %w", err)
+	}
+	defer func() { _ = bfs.Close() }()
 
 	total := len(inv[PageTypeConcept]) + len(inv[PageTypeEntity]) + len(inv[PageTypeSource])
 	var sb strings.Builder
@@ -34,7 +40,7 @@ func RebuildIndex(brainDir, date string) error {
 		label := strings.ToUpper(string(pt)[:1]) + string(pt)[1:]
 		fmt.Fprintf(&sb, "## %s\n\n", label)
 		for _, e := range entries {
-			summary := pageFirstSentence(brainDir, e)
+			summary := pageFirstSentence(bfs, e)
 			if summary != "" {
 				fmt.Fprintf(&sb, "- [[%s|%s]] — %s\n", e.Slug, e.Title, summary)
 			} else {
@@ -48,9 +54,10 @@ func RebuildIndex(brainDir, date string) error {
 	return os.WriteFile(dest, []byte(sb.String()), 0o644)
 }
 
-func pageFirstSentence(brainDir string, e Entry) string {
+// pageFirstSentence reads the page through the brain root, regular files
-	path := filepath.Join(brainDir, "wiki", string(e.Type), e.Slug+".md")
+// only (hyperguild#157).
-	content, err := os.ReadFile(path)
+func pageFirstSentence(bfs *brainfs.Brain, e Entry) string {
+	content, err := bfs.ReadRegular("wiki/" + string(e.Type) + "/" + e.Slug + ".md")
 	if err != nil {
 		return ""
 	}

ingestion/internal/wiki/inventory.go

@@ -3,36 +3,45 @@ package wiki
 
 import (
 	"bufio"
+	"bytes"
+	"errors"
 	"fmt"
-	"os"
+	"io/fs"
-	"path/filepath"
 	"strings"
+
+	"github.com/mathiasbq/hyperguild/ingestion/internal/brainfs"
 )
 
 // LoadInventory walks brain/wiki/ and returns all pages grouped by type.
-// Missing subdirectories are silently skipped.
+// Missing subdirectories are silently skipped. Pages are listed and read
+// through the brain root, regular files only (hyperguild#157); a type
+// directory that is a symlink is an error.
 func LoadInventory(brainDir string) (map[PageType][]Entry, error) {
 	result := map[PageType][]Entry{
 		PageTypeConcept: {},
 		PageTypeEntity:  {},
 		PageTypeSource:  {},
 	}
+	bfs, err := brainfs.Open(brainDir)
+	if err != nil {
+		return nil, fmt.Errorf("open brain: %w", err)
+	}
+	defer func() { _ = bfs.Close() }()
 	for pt := range result {
-		dir := filepath.Join(brainDir, "wiki", string(pt))
+		dir := "wiki/" + string(pt)
-		entries, err := os.ReadDir(dir)
+		names, err := bfs.ReadDirRegular(dir)
-		if os.IsNotExist(err) {
+		if errors.Is(err, fs.ErrNotExist) {
 			continue
 		}
 		if err != nil {
 			return nil, fmt.Errorf("read dir %s: %w", dir, err)
 		}
-		for _, e := range entries {
+		for _, name := range names {
-			if e.IsDir() || !strings.HasSuffix(e.Name(), ".md") {
+			if !strings.HasSuffix(name, ".md") {
 				continue
 			}
-			slug := strings.TrimSuffix(e.Name(), ".md")
+			slug := strings.TrimSuffix(name, ".md")
-			path := filepath.Join(dir, e.Name())
+			title, aliases := readFrontmatter(bfs, dir+"/"+name, slug)
-			title, aliases := readFrontmatter(path, slug)
 			result[pt] = append(result[pt], Entry{Slug: slug, Title: title, Aliases: aliases, Type: pt})
 		}
 	}
@@ -41,15 +50,14 @@ func LoadInventory(brainDir string) (map[PageType][]Entry, error) {
 
 // readFrontmatter extracts title and aliases from YAML frontmatter.
 // Falls back to slug for title and empty aliases on any error.
-func readFrontmatter(path, fallbackSlug string) (title string, aliases []string) {
+func readFrontmatter(bfs *brainfs.Brain, rel, fallbackSlug string) (title string, aliases []string) {
 	title = fallbackSlug
-	f, err := os.Open(path)
+	content, err := bfs.ReadRegular(rel)
 	if err != nil {
 		return
 	}
-	defer f.Close() //nolint:errcheck
 
-	scanner := bufio.NewScanner(f)
+	scanner := bufio.NewScanner(bytes.NewReader(content))
 	inFM := false
 	inAliases := false
 	for scanner.Scan() {

ingestion/internal/wiki/root_test.go

@@ -0,0 +1,79 @@
+package wiki
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/mathiasbq/hyperguild/ingestion/internal/brainfs"
+	"github.com/mathiasbq/hyperguild/ingestion/internal/brainfs/brainfstest"
+)
+
+// hyperguild#157 slice 4: LoadInventory listed wiki/<type>/ with os.ReadDir
+// and opened each .md by path, so a symlinked page put the target's title and
+// aliases into the inventory (and the LLM prompts built from it), and a FIFO
+// hung it. RebuildIndex read first sentences the same way.
+func TestLoadInventoryRefusesSymlinksAndSurvivesSwapRaces(t *testing.T) {
+	brain := t.TempDir()
+	dir := filepath.Join(brain, "wiki", "concepts")
+	require.NoError(t, os.MkdirAll(dir, 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(dir, "ok.md"), []byte("---\ntitle: zqinside\n---\n"), 0o644))
+	require.NoError(t, os.Symlink(brainfstest.Outside(t), filepath.Join(dir, "leak.md")))
+	inBrain := brainfstest.WriteSentinelAt(t, filepath.Join(brain, "raw", "s.md"))
+	require.NoError(t, os.Symlink(filepath.Join("..", "..", "raw", "s.md"), filepath.Join(dir, "inlink.md")))
+
+	inv, err := LoadInventory(brain)
+	require.NoError(t, err)
+	got := fmt.Sprintf("%+v", inv)
+	assert.Contains(t, got, "zqinside", "positive control")
+	assert.NotContains(t, got, brainfstest.Sentinel)
+	assert.NotContains(t, got, "leak", "a symlinked page is not listed")
+	assert.NotContains(t, got, "inlink", "a page linked to another brain file is not listed")
+
+	p := filepath.Join(dir, "r.md")
+	read := func() string {
+		inv, _ := LoadInventory(brain)
+		return fmt.Sprintf("%+v", inv)
+	}
+	brainfstest.AssertNoLeak(t, p, "---\ntitle: zqinside\n---\n", 1500*time.Millisecond, read)
+	brainfstest.AssertNoLeakFrom(t, p, "---\ntitle: zqinside\n---\n", inBrain, 1500*time.Millisecond, read)
+	brainfstest.AssertNoHang(t, p, "---\ntitle: zqinside\n---\n", 1500*time.Millisecond, 6*time.Second, func() { _ = read() })
+}
+
+func TestLoadInventoryReturnsErrNotRegularForASymlinkedTypeDir(t *testing.T) {
+	brain := t.TempDir()
+	require.NoError(t, os.MkdirAll(filepath.Join(brain, "wiki"), 0o755))
+	require.NoError(t, os.MkdirAll(filepath.Join(brain, "raw"), 0o755))
+	require.NoError(t, os.WriteFile(filepath.Join(brain, "raw", "r.md"), []byte("---\ntitle: zqrawonly\n---\n"), 0o644))
+	require.NoError(t, os.Symlink(filepath.Join("..", "raw"), filepath.Join(brain, "wiki", "concepts")))
+	inv, err := LoadInventory(brain)
+	require.ErrorIs(t, err, brainfs.ErrNotRegular)
+	assert.NotContains(t, fmt.Sprintf("%+v", inv), "zqrawonly")
+}
+
+func TestRebuildIndexDoesNotReadAPageSwappedForASymlink(t *testing.T) {
+	brain := t.TempDir()
+	dir := filepath.Join(brain, "wiki", "concepts")
+	require.NoError(t, os.MkdirAll(dir, 0o755))
+	p := filepath.Join(dir, "r.md")
+	require.NoError(t, os.WriteFile(p, []byte("---\ntitle: R\n---\nzqinside sentence.\n"), 0o644))
+	require.NoError(t, RebuildIndex(brain, "2026-10-04"))
+	idx, err := os.ReadFile(filepath.Join(brain, "wiki", "index.md"))
+	require.NoError(t, err)
+	require.Contains(t, string(idx), "zqinside", "positive control: the summary is read")
+
+	read := func() string {
+		_ = RebuildIndex(brain, "2026-10-04")
+		b, _ := os.ReadFile(filepath.Join(brain, "wiki", "index.md"))
+		return string(b)
+	}
+	brainfstest.AssertNoLeak(t, p, "---\ntitle: R\n---\nzqinside sentence.\n", 1500*time.Millisecond, read)
+	inBrain := brainfstest.WriteSentinelAt(t, filepath.Join(brain, "raw", "s.md"))
+	brainfstest.AssertNoLeakFrom(t, p, "---\ntitle: R\n---\nzqinside sentence.\n", inBrain, 1500*time.Millisecond, read)
+	brainfstest.AssertNoHang(t, p, "---\ntitle: R\n---\nzqinside sentence.\n", 1500*time.Millisecond, 6*time.Second, func() { _ = read() })
+}