From e433dcf0c932bf6aa0e4c7e177b93e2c08fc66c0 Mon Sep 17 00:00:00 2001
From: Jakub Zych <jakub@keios.eu>
Date: Wed, 30 Sep 2026 21:21:56 +0200
Subject: [PATCH] feat(11.1-01): publish every module README as an API
 reference page

- discover modules/<m> with non-test Go files; a missing README is a readme: problem
- one GitHub-compatible slug parser.IDs for heading anchors, passed per page
- rewrite links to .md pages and module READMEs to site .html and .md URLs
- search-index.json gains one entry per H2 with 300-char plain text
- add the api section to docs/site.yaml; docsite.Pages exposes reading order
- tests: TestSlugIDs, TestReadmeIngestion, TestEveryModuleInSidebar, TestDocsAIOutputsInSync
---
 cmd/summer/docs_test.go          | 166 ++++++++++++++++
 docs/site.yaml                   |   2 +
 internal/docsite/docsite_test.go | 149 +++++++++++++++
 internal/docsite/emit.go         |  55 ++++--
 internal/docsite/load.go         | 167 +++++++++++++++--
 internal/docsite/render.go       | 312 ++++++++++++++++++++++++++++++-
 6 files changed, 815 insertions(+), 36 deletions(-)
 create mode 100644 internal/docsite/docsite_test.go

diff --git a/cmd/summer/docs_test.go b/cmd/summer/docs_test.go
index 70f8e62..956478a 100644
--- a/cmd/summer/docs_test.go
+++ b/cmd/summer/docs_test.go
@@ -1,10 +1,14 @@
 package main
 
 import (
+	"bufio"
 	"bytes"
+	"io/fs"
 	"os"
 	"path/filepath"
 	"regexp"
+	"slices"
+	"strings"
 	"testing"
 
 	"git.golem15.com/golem15/summercms/internal/docsite"
@@ -57,3 +61,165 @@ func TestDocsBuildRealTree(t *testing.T) {
 		}
 	}
 }
+
+// frameworkModules lists modules/<m> directories that hold a non-test Go
+// file, discovered independently of docsite.
+func frameworkModules(t *testing.T) []string {
+	t.Helper()
+	entries, err := os.ReadDir(filepath.Join(repoRoot, "modules"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	var names []string
+	for _, e := range entries {
+		if !e.IsDir() {
+			continue
+		}
+		goFiles, err := filepath.Glob(filepath.Join(repoRoot, "modules", e.Name(), "*.go"))
+		if err != nil {
+			t.Fatal(err)
+		}
+		if slices.ContainsFunc(goFiles, func(f string) bool { return !strings.HasSuffix(f, "_test.go") }) {
+			names = append(names, e.Name())
+		}
+	}
+	if len(names) == 0 {
+		t.Fatal("no framework modules found")
+	}
+	return names
+}
+
+func TestEveryModuleInSidebar(t *testing.T) {
+	out, _ := buildRealTree(t)
+	index, err := os.ReadFile(filepath.Join(out, "index.html"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	start := bytes.Index(index, []byte(`<nav class="sidebar" aria-label="Documentation">`))
+	end := bytes.Index(index[max(start, 0):], []byte("</nav>"))
+	if start < 0 || end < 0 {
+		t.Fatal("index.html has no sidebar nav")
+	}
+	sidebar := string(index[start : start+end])
+	for _, m := range frameworkModules(t) {
+		if _, err := os.Stat(filepath.Join(out, "api", m+".html")); err != nil {
+			t.Errorf("module %s has no API page: %v", m, err)
+		}
+		if !strings.Contains(sidebar, `href="/api/`+m+`.html"`) {
+			t.Errorf("sidebar does not link api/%s.html", m)
+		}
+	}
+}
+
+// TestDocsAIOutputsInSync asserts that the page tree, the .html pages, the
+// .md siblings, llms.txt and llms-full.txt all list the same pages in the
+// same reading order.
+func TestDocsAIOutputsInSync(t *testing.T) {
+	pages, problems, err := docsite.Pages(docsite.Options{Root: repoRoot})
+	if err != nil || len(problems) > 0 {
+		t.Fatalf("Pages: %v %v", err, problems)
+	}
+	var want []string
+	for _, p := range pages {
+		want = append(want, p.URL)
+	}
+	out, _ := buildRealTree(t)
+
+	var htmlFiles, mdFiles []string
+	if err := filepath.WalkDir(out, func(p string, d fs.DirEntry, err error) error {
+		if err != nil || d.IsDir() {
+			return err
+		}
+		rel, _ := filepath.Rel(out, p)
+		rel = filepath.ToSlash(rel)
+		if strings.HasPrefix(rel, "assets/") {
+			return nil
+		}
+		switch filepath.Ext(rel) {
+		case ".html":
+			htmlFiles = append(htmlFiles, strings.TrimSuffix(rel, ".html"))
+		case ".md":
+			mdFiles = append(mdFiles, strings.TrimSuffix(rel, ".md"))
+		}
+		return nil
+	}); err != nil {
+		t.Fatal(err)
+	}
+	sorted := slices.Sorted(slices.Values(want))
+	slices.Sort(htmlFiles)
+	slices.Sort(mdFiles)
+	if !slices.Equal(htmlFiles, sorted) {
+		t.Errorf(".html pages = %v, want %v", htmlFiles, sorted)
+	}
+	if !slices.Equal(mdFiles, sorted) {
+		t.Errorf(".md pages = %v, want %v", mdFiles, sorted)
+	}
+
+	llms := readLines(t, filepath.Join(out, "llms.txt"))
+	if len(llms) == 0 || llms[0] != "# SummerCMS" {
+		t.Fatalf("llms.txt line 1 = %q, want # SummerCMS", first(llms))
+	}
+	if next := nextNonEmpty(llms, 1); next < 0 || !strings.HasPrefix(llms[next], "> ") {
+		t.Errorf("llms.txt: the line after the H1 must be a > summary")
+	}
+	item := regexp.MustCompile(`^- \[[^\]]+\]\(/([^)]+)\.md\): \S`)
+	var linked []string
+	for i, line := range llms {
+		if strings.HasPrefix(line, "## ") {
+			if next := nextNonEmpty(llms, i+1); next < 0 || !strings.HasPrefix(llms[next], "- [") {
+				t.Errorf("llms.txt: %q is not followed by a link list", line)
+			}
+		}
+		if m := item.FindStringSubmatch(line); m != nil {
+			linked = append(linked, m[1])
+		}
+	}
+	if !slices.Equal(linked, want) {
+		t.Errorf("llms.txt links = %v, want %v", linked, want)
+	}
+
+	var sources []string
+	for _, line := range readLines(t, filepath.Join(out, "llms-full.txt")) {
+		if rest, ok := strings.CutPrefix(line, "Source: /"); ok {
+			sources = append(sources, strings.TrimSuffix(rest, ".html"))
+		}
+	}
+	if !slices.Equal(sources, want) {
+		t.Errorf("llms-full.txt sources = %v, want %v", sources, want)
+	}
+}
+
+func readLines(t *testing.T, path string) []string {
+	t.Helper()
+	f, err := os.Open(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer f.Close()
+	var lines []string
+	sc := bufio.NewScanner(f)
+	sc.Buffer(make([]byte, 0, 1<<20), 1<<24)
+	for sc.Scan() {
+		lines = append(lines, sc.Text())
+	}
+	if err := sc.Err(); err != nil {
+		t.Fatal(err)
+	}
+	return lines
+}
+
+func nextNonEmpty(lines []string, from int) int {
+	for i := from; i < len(lines); i++ {
+		if strings.TrimSpace(lines[i]) != "" {
+			return i
+		}
+	}
+	return -1
+}
+
+func first(lines []string) string {
+	if len(lines) == 0 {
+		return ""
+	}
+	return lines[0]
+}
diff --git a/docs/site.yaml b/docs/site.yaml
index e98929b..f9bed85 100644
--- a/docs/site.yaml
+++ b/docs/site.yaml
@@ -13,3 +13,5 @@ llms_notes:
 sections:
   - name: setup
     title: Setup
+  - name: api
+    title: API reference
diff --git a/internal/docsite/docsite_test.go b/internal/docsite/docsite_test.go
new file mode 100644
index 0000000..7aaeb9f
--- /dev/null
+++ b/internal/docsite/docsite_test.go
@@ -0,0 +1,149 @@
+package docsite
+
+import (
+	"os"
+	"path/filepath"
+	"slices"
+	"strconv"
+	"strings"
+	"testing"
+)
+
+// writeTree writes files (path relative to root, forward slashes) under a
+// fresh temp dir and returns it.
+func writeTree(t *testing.T, files map[string]string) string {
+	t.Helper()
+	root := t.TempDir()
+	for name, body := range files {
+		p := filepath.Join(root, filepath.FromSlash(name))
+		if err := os.MkdirAll(filepath.Dir(p), 0o755); err != nil {
+			t.Fatal(err)
+		}
+		if err := os.WriteFile(p, []byte(body), 0o644); err != nil {
+			t.Fatal(err)
+		}
+	}
+	return root
+}
+
+func problemLines(problems []Problem) []string {
+	var out []string
+	for _, p := range problems {
+		out = append(out, p.String())
+	}
+	return out
+}
+
+const fixtureSite = `title: Acme
+description: Acme docs.
+sections:
+  - name: setup
+    title: Setup
+  - name: api
+    title: API reference
+`
+
+func page(title, section string, order int, body string) string {
+	return "---\ntitle: " + title + "\ndescription: " + title + " page.\nsection: " + section +
+		"\norder: " + strconv.Itoa(order) + "\n---\n# " + title + "\n\n" + body
+}
+
+func TestSlugIDs(t *testing.T) {
+	ids := newSlugIDs()
+	for _, tc := range []struct{ in, want string }{
+		{"Install the CLI", "install-the-cli"},
+		{"Install the CLI", "install-the-cli-1"},
+		{"Install the CLI", "install-the-cli-2"},
+		{"snake_case and dash-case", "snake_case-and-dash-case"},
+		{"What's new? (v2.0)", "whats-new-v20"},
+		{"`bonfire.Call` usage", "bonfirecall-usage"},
+		{"Zażółć gęślą", "zażółć-gęślą"},
+		{"!!!", "section"},
+	} {
+		if got := string(ids.Generate([]byte(tc.in), 0)); got != tc.want {
+			t.Errorf("Generate(%q) = %q, want %q", tc.in, got, tc.want)
+		}
+	}
+	ids.Put([]byte("custom"))
+	if got := string(ids.Generate([]byte("Custom"), 0)); got != "custom-1" {
+		t.Errorf("after Put, Generate(Custom) = %q, want custom-1", got)
+	}
+}
+
+func TestReadmeIngestion(t *testing.T) {
+	files := map[string]string{
+		"docs/site.yaml":         fixtureSite,
+		"docs/index.md":          page("Acme docs", "index", 0, "Read [alpha](../modules/alpha/README.md#usage).\n"),
+		"docs/setup/start.md":    page("Start", "setup", 10, "## First steps\n\nText.\n"),
+		"modules/alpha/alpha.go": "package alpha\n",
+		"modules/alpha/README.md": "# alpha\n\nAlpha does one thing well.\n\n## Usage\n\n" +
+			"See [delta](../delta/README.md#api) and [start](../../docs/setup/start.md).\n",
+		"modules/beta/beta.go":        "package beta\n",
+		"modules/gamma/gamma_test.go": "package gamma\n",
+		"modules/delta/delta.go":      "package delta\n",
+		"modules/delta/README.md":     "# delta\n\nDelta does another thing.\n\n## API\n\nText.\n",
+	}
+	root := writeTree(t, files)
+
+	pages, problems, err := Pages(Options{Root: root})
+	if err != nil {
+		t.Fatal(err)
+	}
+	want := "modules/beta: readme: package has Go files but no README.md"
+	if got := problemLines(problems); !slices.Equal(got, []string{want}) {
+		t.Fatalf("problems = %q, want [%q]", got, want)
+	}
+	var urls []string
+	for _, p := range pages {
+		urls = append(urls, p.URL)
+	}
+	if !slices.Equal(urls, []string{"index", "setup/start", "api/alpha", "api/delta"}) {
+		t.Fatalf("reading order = %v", urls)
+	}
+	alpha := pages[2]
+	if alpha.Title != "alpha" || alpha.Description != "Alpha does one thing well." || alpha.Section != "api" ||
+		alpha.Module != "alpha" || alpha.Source != "modules/alpha/README.md" {
+		t.Fatalf("alpha page = %+v", alpha)
+	}
+
+	if err := os.WriteFile(filepath.Join(root, "modules/beta/README.md"), []byte("# beta\n\nBeta.\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	out := filepath.Join(t.TempDir(), "site")
+	res, problems, err := Build(Options{Root: root, Out: out})
+	if err != nil || len(problems) > 0 {
+		t.Fatalf("Build: %v %q", err, problemLines(problems))
+	}
+	if res.Pages != 5 {
+		t.Fatalf("pages = %d, want 5", res.Pages)
+	}
+	read := func(name string) string {
+		t.Helper()
+		b, err := os.ReadFile(filepath.Join(out, name))
+		if err != nil {
+			t.Fatal(err)
+		}
+		return string(b)
+	}
+	html := read("api/alpha.html")
+	for _, want := range []string{`href="/api/delta.html#api"`, `href="/setup/start.html"`, `<h2 id="usage">Usage</h2>`} {
+		if !strings.Contains(html, want) {
+			t.Errorf("api/alpha.html missing %s", want)
+		}
+	}
+	if strings.Contains(html, "<h1>alpha</h1>\n<p class=\"lead\">Alpha does one thing well.</p>\n<p>Alpha does") {
+		t.Error("summary line rendered twice")
+	}
+	md := read("api/alpha.md")
+	if !strings.HasPrefix(md, "# alpha\n\n> Alpha does one thing well.\n\n## Usage\n") ||
+		!strings.Contains(md, "(/api/delta.md#api)") || !strings.Contains(md, "(/setup/start.md)") {
+		t.Errorf("api/alpha.md = %q", md)
+	}
+	if idx := read("index.html"); !strings.Contains(idx, `href="/api/alpha.html#usage"`) {
+		t.Error("index.html does not link the rewritten module README")
+	}
+	search := read("search-index.json")
+	if !strings.Contains(search, `{"p":2,"a":"usage","h":"Usage","x":"See delta and start."}`) {
+		t.Errorf("search index missing the alpha Usage entry: %s", search)
+	}
+}
diff --git a/internal/docsite/emit.go b/internal/docsite/emit.go
index c72bbdc..807fe28 100644
--- a/internal/docsite/emit.go
+++ b/internal/docsite/emit.go
@@ -77,11 +77,15 @@ func (s *site) render() ([]Problem, error) {
 			return nil, fmt.Errorf("docsite: render template for %s: %w", p.Source, err)
 		}
 		s.outputs[p.URL+".html"] = buf.Bytes()
-		s.outputs[p.URL+".md"] = s.pageMarkdown(p)
+	}
+	bodies := make([]string, len(s.pages))
+	for i, p := range s.pages {
+		bodies[i] = markdownBody(p, rendered[i].mdLinks)
+		s.outputs[p.URL+".md"] = pageMarkdown(p, bodies[i])
 	}
 	s.outputs["llms.txt"] = s.llmsTxt()
-	s.outputs["llms-full.txt"] = s.llmsFull()
-	idx, err := s.searchIndex()
+	s.outputs["llms-full.txt"] = s.llmsFull(bodies)
+	idx, err := s.searchIndex(rendered)
 	if err != nil {
 		return nil, err
 	}
@@ -109,25 +113,27 @@ func (s *site) nav(current *Page) []navSection {
 }
 
 // pageMarkdown returns the clean Markdown sibling of a page: no
-// frontmatter, "# Title", "> description", then the body without its H1.
-func (s *site) pageMarkdown(p *Page) []byte {
+// frontmatter, "# Title", "> description", then the transformed body.
+func pageMarkdown(p *Page, body string) []byte {
 	var b bytes.Buffer
 	fmt.Fprintf(&b, "# %s\n\n> %s\n\n", p.Title, p.Description)
-	b.WriteString(s.markdownBody(p))
+	b.WriteString(body)
 	return b.Bytes()
 }
 
 // markdownBody is the page body without its H1 and leading blank lines,
-// ending in one newline.
-func (s *site) markdownBody(p *Page) string {
+// with links to other pages pointing at their .md URLs, ending in one
+// newline.
+func markdownBody(p *Page, links map[string]string) string {
 	body := string(p.Body)
-	if first, rest, ok := strings.Cut(body, "\n"); ok && strings.HasPrefix(first, "# ") {
-		body = rest
-	} else if !ok && strings.HasPrefix(first, "# ") {
+	if first, rest, ok := strings.Cut(body, "\n"); strings.HasPrefix(first, "# ") {
 		body = ""
+		if ok {
+			body = rest
+		}
 	}
-	body = strings.TrimLeft(body, "\n")
-	return strings.TrimRight(body, "\n") + "\n"
+	body = strings.Trim(body, "\n")
+	return rewriteMarkdown(body, links) + "\n"
 }
 
 // llmsTxt writes the llms.txt index (llmstxt.org shape).
@@ -161,14 +167,14 @@ func (s *site) llmsTxt() []byte {
 }
 
 // llmsFull concatenates every page in reading order.
-func (s *site) llmsFull() []byte {
+func (s *site) llmsFull(bodies []string) []byte {
 	var b bytes.Buffer
 	for i, p := range s.pages {
 		if i > 0 {
 			b.WriteString("\n")
 		}
 		fmt.Fprintf(&b, "# %s\nSource: %s\n\n%s\n\n", p.Title, s.url(p.URL+".html"), p.Description)
-		b.WriteString(s.markdownBody(p))
+		b.WriteString(bodies[i])
 	}
 	return b.Bytes()
 }
@@ -191,14 +197,29 @@ type searchIndex struct {
 	Entries []searchEntry `json:"e"`
 }
 
-func (s *site) searchIndex() ([]byte, error) {
+const searchTextMax = 300
+
+// searchIndex lists every page and one entry per H2 heading, so a hit
+// deep-links to its anchor.
+func (s *site) searchIndex(rendered []renderedPage) ([]byte, error) {
 	idx := searchIndex{Pages: []searchPage{}, Entries: []searchEntry{}}
-	for _, p := range s.pages {
+	for i, p := range s.pages {
 		sec := s.cfg.Title
 		if p.Section != indexSection {
 			sec = s.cfg.sectionTitle(p.Section)
 		}
 		idx.Pages = append(idx.Pages, searchPage{URL: s.url(p.URL + ".html"), Title: p.Title, Section: sec})
+		for _, h := range rendered[i].headings {
+			if h.Level != 2 {
+				continue
+			}
+			idx.Entries = append(idx.Entries, searchEntry{
+				Page:    i,
+				Anchor:  h.ID,
+				Heading: h.Text,
+				Text:    sectionText(h.node, p.Body, searchTextMax),
+			})
+		}
 	}
 	raw, err := json.Marshal(idx)
 	if err != nil {
diff --git a/internal/docsite/load.go b/internal/docsite/load.go
index e339f97..5d07b72 100644
--- a/internal/docsite/load.go
+++ b/internal/docsite/load.go
@@ -3,6 +3,7 @@ package docsite
 import (
 	"bytes"
 	"cmp"
+	"errors"
 	"fmt"
 	"io/fs"
 	"os"
@@ -139,12 +140,13 @@ func (s Site) hasSection(name string) bool {
 
 // site is one assembled documentation site.
 type site struct {
-	opts    Options
-	cfg     Site
-	cfgRaw  []byte
-	base    string
-	pages   []*Page // reading order
-	outputs map[string][]byte
+	opts     Options
+	cfg      Site
+	cfgRaw   []byte
+	base     string
+	pages    []*Page // reading order
+	bySource map[string]*Page
+	outputs  map[string][]byte
 }
 
 // rel returns the display path of an absolute path: repository-relative
@@ -156,13 +158,47 @@ func (s *site) rel(abs string) string {
 	return filepath.ToSlash(abs)
 }
 
+// Pages loads the docs pages and the module READMEs and returns them in
+// reading order (the order of the sidebar, the pager and llms-full.txt),
+// with any load problems. It renders nothing.
+func Pages(opts Options) ([]Page, []Problem, error) {
+	s, problems, err := load(opts)
+	if err != nil || s == nil {
+		return nil, problems, err
+	}
+	out := make([]Page, len(s.pages))
+	for i, p := range s.pages {
+		out[i] = *p
+	}
+	return out, problems, nil
+}
+
 // assemble loads, verifies and renders a site in memory.
 func assemble(opts Options) (*site, []Problem, error) {
+	s, problems, err := load(opts)
+	if err != nil || s == nil {
+		return nil, problems, err
+	}
+	if len(problems) == 0 {
+		rp, err := s.render()
+		if err != nil {
+			return nil, nil, err
+		}
+		problems = append(problems, rp...)
+	}
+	sortProblems(problems)
+	return s, problems, nil
+}
+
+// load reads site.yaml, the docs pages and the module READMEs and puts
+// the pages in reading order. A nil site with problems means site.yaml
+// itself is invalid.
+func load(opts Options) (*site, []Problem, error) {
 	opts, err := opts.normalize()
 	if err != nil {
 		return nil, nil, err
 	}
-	s := &site{opts: opts, outputs: map[string][]byte{}}
+	s := &site{opts: opts, outputs: map[string][]byte{}, bySource: map[string]*Page{}}
 	cfgPath := filepath.Join(opts.Src, "site.yaml")
 	raw, err := os.ReadFile(cfgPath)
 	if err != nil {
@@ -182,19 +218,120 @@ func assemble(opts Options) (*site, []Problem, error) {
 	if err != nil {
 		return nil, nil, err
 	}
-	s.order(guides)
-	problems = append(problems, s.emptySections()...)
-	if len(problems) == 0 {
-		rp, err := s.render()
-		if err != nil {
-			return nil, nil, err
-		}
-		problems = append(problems, rp...)
+	modules, mp, err := s.loadModules()
+	if err != nil {
+		return nil, nil, err
 	}
+	problems = append(problems, mp...)
+	if len(modules) > 0 && !cfg.hasSection(apiSection) {
+		problems = append(problems, Problem{File: s.rel(cfgPath), Line: 1, Rule: "section",
+			Message: fmt.Sprintf("%q is not listed (every module README is published there)", apiSection)})
+	}
+	s.order(append(guides, modules...))
+	for _, p := range s.pages {
+		s.bySource[p.Source] = p
+	}
+	problems = append(problems, s.emptySections()...)
 	sortProblems(problems)
 	return s, problems, nil
 }
 
+// loadModules turns every modules/<name> directory that holds a non-test Go
+// file into an API reference page built from its README.md. The module
+// list is discovered, never hard-coded.
+func (s *site) loadModules() ([]*Page, []Problem, error) {
+	names, err := moduleNames(s.opts.Root)
+	if err != nil {
+		return nil, nil, err
+	}
+	var pages []*Page
+	var problems []Problem
+	for i, name := range names {
+		dir := filepath.Join(s.opts.Root, "modules", name)
+		readme := filepath.Join(dir, "README.md")
+		raw, err := os.ReadFile(readme)
+		if errors.Is(err, fs.ErrNotExist) {
+			problems = append(problems, Problem{File: s.rel(dir), Rule: "readme", Message: "package has Go files but no README.md"})
+			continue
+		}
+		if err != nil {
+			return nil, nil, fmt.Errorf("docsite: read %s: %w", readme, err)
+		}
+		title, desc, body, ok := splitReadme(raw)
+		if !ok {
+			problems = append(problems, Problem{File: s.rel(readme), Line: 1, Rule: "readme", Message: `first line must be the "# <module>" title followed by a summary line`})
+			continue
+		}
+		pages = append(pages, &Page{
+			Source:      s.rel(readme),
+			URL:         apiSection + "/" + name,
+			Section:     apiSection,
+			Title:       title,
+			Description: desc,
+			Order:       i,
+			Module:      name,
+			Body:        body,
+			BodyLine:    1,
+			abs:         readme,
+		})
+	}
+	return pages, problems, nil
+}
+
+// moduleNames lists the top-level modules/ directories that hold a non-test
+// Go file, sorted by name.
+func moduleNames(root string) ([]string, error) {
+	entries, err := os.ReadDir(filepath.Join(root, "modules"))
+	if errors.Is(err, fs.ErrNotExist) {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, fmt.Errorf("docsite: read modules: %w", err)
+	}
+	var names []string
+	for _, e := range entries {
+		if !e.IsDir() || strings.HasPrefix(e.Name(), ".") || strings.HasPrefix(e.Name(), "_") {
+			continue
+		}
+		files, err := os.ReadDir(filepath.Join(root, "modules", e.Name()))
+		if err != nil {
+			return nil, fmt.Errorf("docsite: read module %s: %w", e.Name(), err)
+		}
+		if slices.ContainsFunc(files, func(f fs.DirEntry) bool {
+			n := f.Name()
+			return !f.IsDir() && strings.HasSuffix(n, ".go") && !strings.HasSuffix(n, "_test.go")
+		}) {
+			names = append(names, e.Name())
+		}
+	}
+	slices.Sort(names)
+	return names, nil
+}
+
+// splitReadme returns a README's H1 text, its summary line and the body
+// with the summary line blanked (line numbers are kept so problems still
+// cite the README's own lines).
+func splitReadme(raw []byte) (title, desc string, body []byte, ok bool) {
+	lines := strings.Split(string(raw), "\n")
+	if len(lines) == 0 || !strings.HasPrefix(lines[0], "# ") {
+		return "", "", nil, false
+	}
+	title = strings.TrimSpace(strings.TrimPrefix(lines[0], "# "))
+	for i := 1; i < len(lines); i++ {
+		line := strings.TrimSpace(lines[i])
+		if line == "" {
+			continue
+		}
+		if strings.HasPrefix(line, "#") || title == "" {
+			return "", "", nil, false
+		}
+		desc = line
+		lines[i] = ""
+		return title, desc, []byte(strings.Join(lines, "\n")), true
+	}
+	return "", "", nil, false
+}
+
 // loadGuides walks Src and loads every page with its frontmatter.
 func (s *site) loadGuides() ([]*Page, []Problem, error) {
 	files, err := walkPages(s.opts.Src)
diff --git a/internal/docsite/render.go b/internal/docsite/render.go
index 822d747..b9f5ac8 100644
--- a/internal/docsite/render.go
+++ b/internal/docsite/render.go
@@ -2,7 +2,13 @@ package docsite
 
 import (
 	"bytes"
+	"cmp"
 	"fmt"
+	"path"
+	"slices"
+	"strings"
+	"unicode"
+	"unicode/utf8"
 
 	"github.com/yuin/goldmark"
 	"github.com/yuin/goldmark/ast"
@@ -23,6 +29,7 @@ func newMarkdown() goldmark.Markdown {
 			parser.WithAutoHeadingID(),
 			parser.WithASTTransformers(
 				util.Prioritized(h1Stripper{}, 100),
+				util.Prioritized(linkRewriter{}, 200),
 			),
 		),
 	)
@@ -38,18 +45,315 @@ func (h1Stripper) Transform(doc *ast.Document, _ text.Reader, _ parser.Context)
 	}
 }
 
+// slugIDs is the one heading-ID algorithm of the site, shared by the
+// renderer, the search index and the link checker. It is GitHub-compatible:
+// lowercase, Unicode letters, digits, "_" and "-" kept, spaces mapped to
+// "-", other punctuation dropped, and duplicates suffixed -1, -2.
+type slugIDs struct {
+	used map[string]bool
+}
+
+func newSlugIDs() *slugIDs {
+	return &slugIDs{used: map[string]bool{}}
+}
+
+// Generate implements parser.IDs.
+func (s *slugIDs) Generate(value []byte, _ ast.NodeKind) []byte {
+	base := slugify(string(value))
+	if base == "" {
+		base = "section"
+	}
+	id := base
+	for i := 1; s.used[id]; i++ {
+		id = fmt.Sprintf("%s-%d", base, i)
+	}
+	s.used[id] = true
+	return []byte(id)
+}
+
+// Put implements parser.IDs.
+func (s *slugIDs) Put(value []byte) {
+	s.used[string(value)] = true
+}
+
+func slugify(v string) string {
+	var b strings.Builder
+	for _, r := range strings.ToLower(strings.TrimSpace(v)) {
+		switch {
+		case unicode.IsLetter(r) || unicode.IsDigit(r) || r == '_' || r == '-':
+			b.WriteRune(r)
+		case unicode.IsSpace(r):
+			b.WriteByte('-')
+		}
+	}
+	return b.String()
+}
+
+var pageKey = parser.NewContextKey()
+
+// pageContext carries per-page state into the AST transformers.
+type pageContext struct {
+	site *site
+	page *Page
+	// mdLinks maps a rewritten link destination to its .md form, for the
+	// raw Markdown output.
+	mdLinks map[string]string
+}
+
+// linkRewriter points relative links at other pages to their site URLs:
+// a guide's link to another page's .md, or to modules/<m>/README.md, and a
+// README's ../<m>/README.md all become the target page. Fragments are kept;
+// external and mailto links are untouched.
+type linkRewriter struct{}
+
+func (linkRewriter) Transform(doc *ast.Document, _ text.Reader, pc parser.Context) {
+	pctx, _ := pc.Get(pageKey).(*pageContext)
+	if pctx == nil {
+		return
+	}
+	_ = ast.Walk(doc, func(n ast.Node, entering bool) (ast.WalkStatus, error) {
+		link, ok := n.(*ast.Link)
+		if !entering || !ok {
+			return ast.WalkContinue, nil
+		}
+		dest := string(link.Destination)
+		target, frag, ok := pctx.site.resolveLink(pctx.page, dest)
+		if !ok {
+			return ast.WalkContinue, nil
+		}
+		link.Destination = []byte(pctx.site.url(target.URL+".html") + frag)
+		pctx.mdLinks[dest] = pctx.site.url(target.URL+".md") + frag
+		return ast.WalkContinue, nil
+	})
+}
+
+// resolveLink maps a relative link destination in page from to the page it
+// names, with its "#fragment" (or "").
+func (s *site) resolveLink(from *Page, dest string) (*Page, string, bool) {
+	if dest == "" || strings.HasPrefix(dest, "#") || strings.HasPrefix(dest, "/") || hasScheme(dest) {
+		return nil, "", false
+	}
+	target, frag, _ := strings.Cut(dest, "#")
+	if frag != "" {
+		frag = "#" + frag
+	}
+	if !strings.HasSuffix(target, ".md") {
+		return nil, "", false
+	}
+	p, ok := s.bySource[path.Clean(path.Join(path.Dir(from.Source), target))]
+	if !ok {
+		return nil, "", false
+	}
+	return p, frag, true
+}
+
+func hasScheme(dest string) bool {
+	i := strings.IndexByte(dest, ':')
+	return i > 0 && !strings.ContainsAny(dest[:i], "/?#")
+}
+
+// heading is one heading of a rendered page.
+type heading struct {
+	Level int
+	ID    string
+	Text  string
+	node  *ast.Heading
+}
+
 // renderedPage is one page after parsing and rendering.
 type renderedPage struct {
-	html []byte
+	html     []byte
+	doc      ast.Node
+	headings []heading
+	mdLinks  map[string]string
+}
+
+// parsePage parses a page body with a fresh slug-ID table and the page's
+// transformer context.
+func (s *site) parsePage(md goldmark.Markdown, p *Page) (ast.Node, *pageContext) {
+	pctx := &pageContext{site: s, page: p, mdLinks: map[string]string{}}
+	ctx := parser.NewContext(parser.WithIDs(newSlugIDs()))
+	ctx.Set(pageKey, pctx)
+	doc := md.Parser().Parse(text.NewReader(p.Body), parser.WithContext(ctx))
+	return doc, pctx
 }
 
 // renderPage parses and renders a page body to HTML.
 func (s *site) renderPage(md goldmark.Markdown, p *Page) (renderedPage, error) {
-	ctx := parser.NewContext()
-	doc := md.Parser().Parse(text.NewReader(p.Body), parser.WithContext(ctx))
+	doc, pctx := s.parsePage(md, p)
 	var buf bytes.Buffer
 	if err := md.Renderer().Render(&buf, p.Body, doc); err != nil {
 		return renderedPage{}, fmt.Errorf("docsite: render %s: %w", p.Source, err)
 	}
-	return renderedPage{html: buf.Bytes()}, nil
+	return renderedPage{html: buf.Bytes(), doc: doc, headings: pageHeadings(doc, p.Body), mdLinks: pctx.mdLinks}, nil
+}
+
+// headingIDs returns the heading IDs of a page, in document order, from the
+// same parse and slug algorithm the renderer uses.
+func (s *site) headingIDs(md goldmark.Markdown, p *Page) []string {
+	doc, _ := s.parsePage(md, p)
+	var ids []string
+	for _, h := range pageHeadings(doc, p.Body) {
+		ids = append(ids, h.ID)
+	}
+	return ids
+}
+
+func pageHeadings(doc ast.Node, src []byte) []heading {
+	var out []heading
+	_ = ast.Walk(doc, func(n ast.Node, entering bool) (ast.WalkStatus, error) {
+		h, ok := n.(*ast.Heading)
+		if !entering || !ok {
+			return ast.WalkContinue, nil
+		}
+		id, _ := h.AttributeString("id")
+		idb, _ := id.([]byte)
+		out = append(out, heading{Level: h.Level, ID: string(idb), Text: plainText(h, src), node: h})
+		return ast.WalkSkipChildren, nil
+	})
+	return out
+}
+
+// plainText extracts the text of a node from its text nodes only (never
+// from rendered HTML), with whitespace collapsed.
+func plainText(n ast.Node, src []byte) string {
+	var b strings.Builder
+	_ = ast.Walk(n, func(c ast.Node, entering bool) (ast.WalkStatus, error) {
+		if !entering {
+			return ast.WalkContinue, nil
+		}
+		switch t := c.(type) {
+		case *ast.Text:
+			b.Write(t.Segment.Value(src))
+			if t.SoftLineBreak() || t.HardLineBreak() {
+				b.WriteByte(' ')
+			}
+		case *ast.String:
+			b.Write(t.Value)
+		case *ast.FencedCodeBlock, *ast.CodeBlock, *ast.HTMLBlock, *ast.RawHTML:
+			return ast.WalkSkipChildren, nil
+		case *ast.Paragraph, *ast.Heading, *ast.ListItem, *ast.TextBlock:
+			b.WriteByte(' ')
+		}
+		return ast.WalkContinue, nil
+	})
+	return strings.Join(strings.Fields(b.String()), " ")
+}
+
+// sectionText is the plain text of the blocks after h up to the next
+// heading of the same or a higher level, cut to max runes.
+func sectionText(h *ast.Heading, src []byte, max int) string {
+	var parts []string
+	for n := h.NextSibling(); n != nil; n = n.NextSibling() {
+		if next, ok := n.(*ast.Heading); ok && next.Level <= h.Level {
+			break
+		}
+		if t := plainText(n, src); t != "" {
+			parts = append(parts, t)
+		}
+	}
+	return truncateRunes(strings.Join(parts, " "), max)
+}
+
+func truncateRunes(s string, max int) string {
+	if utf8.RuneCountInString(s) <= max {
+		return s
+	}
+	r := []rune(s)
+	return strings.TrimSpace(string(r[:max]))
+}
+
+// fence is one fenced code block found by scanning Markdown lines.
+type fence struct {
+	open, close int // 0-based line indexes; close == -1 when unterminated
+	indent      int
+	char        byte
+	count       int
+	info        string
+}
+
+// scanFences finds fenced code blocks in Markdown lines, following the
+// CommonMark opening and closing rules.
+func scanFences(lines []string) []fence {
+	var out []fence
+	for i := 0; i < len(lines); i++ {
+		f, ok := openFence(lines[i])
+		if !ok {
+			continue
+		}
+		f.open, f.close = i, -1
+		for j := i + 1; j < len(lines); j++ {
+			if closesFence(lines[j], f) {
+				f.close = j
+				break
+			}
+		}
+		out = append(out, f)
+		if f.close < 0 {
+			break
+		}
+		i = f.close
+	}
+	return out
+}
+
+func openFence(line string) (fence, bool) {
+	trimmed := strings.TrimLeft(line, " ")
+	if len(trimmed) < 3 || (trimmed[0] != '`' && trimmed[0] != '~') {
+		return fence{}, false
+	}
+	c := trimmed[0]
+	n := 0
+	for n < len(trimmed) && trimmed[n] == c {
+		n++
+	}
+	if n < 3 {
+		return fence{}, false
+	}
+	info := strings.TrimSpace(trimmed[n:])
+	if c == '`' && strings.ContainsRune(info, '`') {
+		return fence{}, false
+	}
+	return fence{indent: len(line) - len(trimmed), char: c, count: n, info: info}, true
+}
+
+func closesFence(line string, f fence) bool {
+	trimmed := strings.TrimLeft(line, " ")
+	n := 0
+	for n < len(trimmed) && trimmed[n] == f.char {
+		n++
+	}
+	return n >= f.count && strings.TrimSpace(trimmed[n:]) == ""
+}
+
+// rewriteMarkdown applies the raw-output transforms to a Markdown body:
+// link destinations outside code fences are replaced through links.
+func rewriteMarkdown(body string, links map[string]string) string {
+	lines := strings.Split(body, "\n")
+	inFence := make([]bool, len(lines))
+	for _, f := range scanFences(lines) {
+		end := f.close
+		if end < 0 {
+			end = len(lines) - 1
+		}
+		for i := f.open; i <= end; i++ {
+			inFence[i] = true
+		}
+	}
+	origs := make([]string, 0, len(links))
+	for o := range links {
+		origs = append(origs, o)
+	}
+	slices.SortFunc(origs, func(a, b string) int { return cmp.Or(cmp.Compare(len(b), len(a)), strings.Compare(a, b)) })
+	for i, line := range lines {
+		if inFence[i] || !strings.Contains(line, "](") {
+			continue
+		}
+		for _, o := range origs {
+			line = strings.ReplaceAll(line, "]("+o+")", "]("+links[o]+")")
+			line = strings.ReplaceAll(line, "]("+o+" ", "]("+links[o]+" ")
+		}
+		lines[i] = line
+	}
+	return strings.Join(lines, "\n")
 }