diff --git a/crates/cph-check/src/lib.rs b/crates/cph-check/src/lib.rs
index 7e5a491..d547846 100644
--- a/crates/cph-check/src/lib.rs
+++ b/crates/cph-check/src/lib.rs
@@ -304,6 +304,7 @@ pub fn run_shell_target(root: &Path, engine: &Engine, target: &str) -> ShellRunR
         .filter_map(|s| match s {
             cph_model::Step::Shell { run } => Some(run.as_str()),
             cph_model::Step::TypstCompile { .. } => None,
+            cph_model::Step::AssembleMarkdown { .. } => None,
         })
         .collect();
 
@@ -390,6 +391,384 @@ fn run_one_shell(cwd: &Path, run: &str) -> ShellStepOutcome {
     }
 }
 
+/// One assemble-markdown step's outcome (for [`run_markdown_assemble_target`]).
+#[derive(Debug, Clone, PartialEq)]
+pub struct MarkdownAssembleOutcome {
+    /// The field name that was assembled (verbatim from the step's `field`).
+    pub field: String,
+    /// How many elements' `<field>.md` files were read and concatenated.
+    pub parts_read: usize,
+    /// The assembled markdown body (empty if nothing was read).
+    pub body: String,
+    /// Where the assembled file was written (relative to the engineering root);
+    /// `None` if the write failed or was skipped.
+    pub written: Option<String>,
+    /// `None` on success; an error message if reading a required file or writing
+    /// the artifact failed.
+    pub error: Option<String>,
+}
+
+impl MarkdownAssembleOutcome {
+    /// Whether the step succeeded (wrote the artifact with no error).
+    pub fn ok(&self) -> bool {
+        self.error.is_none() && self.written.is_some()
+    }
+}
+
+/// The result of [`run_markdown_assemble_target`].
+#[derive(Debug, Clone, PartialEq)]
+pub struct MarkdownAssembleReport {
+    /// The check phases' report (load → structural → schema). Assembly runs only
+    /// when this has no errors.
+    pub check: CheckReport,
+    /// Each assemble-markdown step's outcome, in declared order. Empty if the
+    /// build was refused (check errors, unknown target, no assemble steps).
+    pub outcomes: Vec<MarkdownAssembleOutcome>,
+    /// `true` if the target was found, check passed, and every assemble step wrote
+    /// its artifact without error.
+    pub ok: bool,
+}
+
+/// Execute the `AssembleMarkdown` steps of a target (ADR-0015): concatenate each
+/// element's `<field>.md` markdown content file in `[[parts]]` order into the
+/// target's single-file artifact. This is the **third typed step**: unlike
+/// [`build`] (typst template → PDF) the framework owns the read/concatenate/write
+/// itself (not a typst compile, not an external tool like [`run_shell_target`]).
+///
+/// Contract / safety (same three boundaries as the shell step, ADR-0013/0015):
+/// - Runs the check phases first (load → structural → schema). Any error refuses
+///   the run (`ok == false`, nothing written) — assembling a broken lesson is
+///   not attempted.
+/// - Walks `lesson.parts` **in declared order**, reading `<part.path>/<field>.md`
+///   when it exists (skipping absent ones — these fields are optional), joining
+///   them with a blank line, and writing to the artifact's single-file
+///   `filepath` (relative to the engineering root). Each `slides.md`/etc.
+///   authors its own heading; the assembler injects nothing (presentation in the
+///   content, ADR-0011).
+/// - **Opt-in by construction**: only reached from `cph build --target <name>`,
+///   never from `check`. A non-typst target, so `check`'s compile phase skips it.
+/// - A write error is a **build-process error**, not a lesson diagnostic — it is
+///   reported in `MarkdownAssembleOutcome::error`, not the 6-class taxonomy.
+pub fn run_markdown_assemble_target(
+    root: &Path,
+    engine: &Engine,
+    target: &str,
+) -> MarkdownAssembleReport {
+    let mut diags = Vec::new();
+
+    let (lesson, load_diags) = cph_model::load(root);
+    diags.extend(load_diags);
+
+    let Some(lesson) = lesson else {
+        return MarkdownAssembleReport {
+            check: CheckReport {
+                diagnostics: dedup(diags),
+                lesson_loaded: false,
+            },
+            outcomes: Vec::new(),
+            ok: false,
+        };
+    };
+
+    let known = cph_schema::known_kinds();
+    run_structural_and_schema(&lesson, known, &mut diags);
+
+    // Signature uniform with `build`/`run_shell_target`; the engine is not used
+    // to assemble markdown (no typst compile), but callers pass it for API
+    // consistency.
+    let _ = engine;
+
+    let has_error = diags.iter().any(|d| d.severity == Severity::Error);
+
+    let Some(tc) = lesson.targets.iter().find(|t| t.name == target) else {
+        diags.push(
+            Diagnostic::error(
+                DiagCode::SchemaViolation,
+                format!("target '{target}' not declared in manifest"),
+            )
+            .with_hint(format!(
+                "add a `[targets.{target}]` table, or run a declared target"
+            )),
+        );
+        return MarkdownAssembleReport {
+            check: CheckReport {
+                diagnostics: dedup(diags),
+                lesson_loaded: true,
+            },
+            outcomes: Vec::new(),
+            ok: false,
+        };
+    };
+
+    let assemble_steps: Vec<&str> = tc
+        .steps
+        .iter()
+        .filter_map(|s| match s {
+            cph_model::Step::AssembleMarkdown { field } => Some(field.as_str()),
+            _ => None,
+        })
+        .collect();
+
+    if has_error || assemble_steps.is_empty() {
+        return MarkdownAssembleReport {
+            check: CheckReport {
+                diagnostics: dedup(diags),
+                lesson_loaded: true,
+            },
+            outcomes: Vec::new(),
+            ok: false,
+        };
+    }
+
+    // Run each assemble-markdown step. First failure stops the sequence.
+    let mut outcomes = Vec::new();
+    let mut all_ok = true;
+    for field in assemble_steps {
+        let outcome = assemble_one(&lesson, field);
+        let ok = outcome.ok();
+        outcomes.push(outcome);
+        if !ok {
+            all_ok = false;
+            break;
+        }
+    }
+
+    MarkdownAssembleReport {
+        check: CheckReport {
+            diagnostics: dedup(diags),
+            lesson_loaded: true,
+        },
+        outcomes,
+        ok: all_ok,
+    }
+}
+
+/// Assemble one markdown `field` across all parts in order, writing the
+/// single-file artifact.
+fn assemble_one(lesson: &cph_model::Lesson, field: &str) -> MarkdownAssembleOutcome {
+    let mut chunks: Vec<String> = Vec::new();
+    let mut parts_read = 0usize;
+
+    // Inject the course-level title as the document's h1. The title is course
+    // metadata (from `[info]` in the manifest), not per-element content, so the
+    // assembler prepends it; each element's `<field>.md` then contributes its
+    // `##` part heading and `###` 小节. (ADR-0011's "presentation lives in the
+    // content" is about per-element presentation; the single course title is a
+    // document-level fact the assembler owns.)
+    if !lesson.info.title.trim().is_empty() {
+        chunks.push(format!("# {}", lesson.info.title.trim()));
+    }
+
+    for part in &lesson.parts {
+        if !part.descriptor.dir.is_dir() {
+            continue;
+        }
+        let md = part.descriptor.dir.join(format!("{field}.md"));
+        if md.is_file() {
+            match std::fs::read_to_string(&md) {
+                Ok(body) => {
+                    chunks.push(body);
+                    parts_read += 1;
+                }
+                Err(e) => {
+                    return MarkdownAssembleOutcome {
+                        field: field.to_string(),
+                        parts_read,
+                        body: String::new(),
+                        written: None,
+                        error: Some(format!("failed to read '{}': {e}", md.display())),
+                    };
+                }
+            }
+        }
+        // Absent `<field>.md` is fine — the field is optional (ADR-0015).
+    }
+
+    let body = chunks.join("\n\n");
+
+    // Resolve the target's single-file artifact path. (The assemble target is a
+    // single-file artifact by construction; a file-tree target has no single
+    // filepath to write to, which is a manifest error surfaced here.)
+    let tc = lesson.targets.iter().find(|t| {
+        t.steps
+            .iter()
+            .any(|s| matches!(s, cph_model::Step::AssembleMarkdown { field: f } if f == field))
+    });
+    let Some(tc) = tc else {
+        return MarkdownAssembleOutcome {
+            field: field.to_string(),
+            parts_read,
+            body,
+            written: None,
+            error: Some("no target carries this assemble-markdown step".into()),
+        };
+    };
+    let cph_model::Artifact::SingleFile { filepath } = &tc.artifact else {
+        return MarkdownAssembleOutcome {
+            field: field.to_string(),
+            parts_read,
+            body,
+            written: None,
+            error: Some(format!(
+                "target '{}' has a non-single-file artifact; assemble-markdown writes a single file",
+                tc.name
+            )),
+        };
+    };
+
+    let out_path = lesson.root.join(filepath);
+    if let Some(parent) = out_path.parent() {
+        if let Err(e) = std::fs::create_dir_all(parent) {
+            return MarkdownAssembleOutcome {
+                field: field.to_string(),
+                parts_read,
+                body,
+                written: None,
+                error: Some(format!("failed to create '{}': {e}", parent.display())),
+            };
+        }
+    }
+    match std::fs::write(&out_path, &body) {
+        Ok(()) => {
+            // Collect referenced image assets so the build dir is self-contained.
+            // The assembled `body` is written under `build_root` (the artifact
+            // filepath's parent); its `![](rel)` image references resolve against
+            // that build root. So each referenced local image is copied from the
+            // engineering root (`lesson.root/<rel>`) into the build root
+            // (`build_root/<rel>`), preserving its relative path. A referenced
+            // image that is missing at the engineering root is a build-process
+            // error (a self-contained build with a dangling ref is broken) — it
+            // does not enter the 6-class diagnostic taxonomy (ADR-0015).
+            if let Some(build_root) = out_path.parent() {
+                if let Err(e) = collect_image_assets(&lesson.root, build_root, &body) {
+                    return MarkdownAssembleOutcome {
+                        field: field.to_string(),
+                        parts_read,
+                        body,
+                        written: Some(filepath.display().to_string()),
+                        error: Some(e),
+                    };
+                }
+            }
+            MarkdownAssembleOutcome {
+                field: field.to_string(),
+                parts_read,
+                body,
+                written: Some(filepath.display().to_string()),
+                error: None,
+            }
+        }
+        Err(e) => MarkdownAssembleOutcome {
+            field: field.to_string(),
+            parts_read,
+            body,
+            written: None,
+            error: Some(format!("failed to write '{}': {e}", out_path.display())),
+        },
+    }
+}
+
+/// Copy each local markdown image reference in `body` from the engineering root
+/// `eng_root` into the build root `build_root`, preserving the reference's
+/// relative path. Returns `Err(message)` on the first referenced image that is
+/// missing at `eng_root` (a build-process error, ADR-0015).
+///
+/// References are `![alt](target)` markdown images; `target` is taken up to the
+/// first whitespace (a possible ` "title"` suffix is stripped). Absolute URLs
+/// (`http://`, `https://`, `data:`) are left untouched — they are not local
+/// assets to collect. Duplicate references copy once.
+fn collect_image_assets(eng_root: &Path, build_root: &Path, body: &str) -> Result<(), String> {
+    let mut seen: std::collections::BTreeSet<String> = std::collections::BTreeSet::new();
+    for target in image_refs(body) {
+        if target.is_empty()
+            || target.starts_with("http://")
+            || target.starts_with("https://")
+            || target.starts_with("data:")
+        {
+            continue;
+        }
+        if !seen.insert(target.to_string()) {
+            continue;
+        }
+        let src = eng_root.join(target);
+        if !src.is_file() {
+            return Err(format!(
+                "slides references image '{}' which is missing from the engineering root",
+                src.display()
+            ));
+        }
+        let dst = build_root.join(target);
+        if let Some(parent) = dst.parent() {
+            if let Err(e) = std::fs::create_dir_all(parent) {
+                return Err(format!("failed to create '{}': {e}", parent.display()));
+            }
+        }
+        if let Err(e) = std::fs::copy(&src, &dst) {
+            return Err(format!("failed to copy '{}' → '{}': {e}", src.display(), dst.display()));
+        }
+    }
+    Ok(())
+}
+
+/// Scan `body` for markdown image references `![alt](target)` and yield each
+/// `target` (whitespace-trimmed of a trailing ` "title"`). Hand-rolled to avoid
+/// pulling a regex dependency for a single, simple pattern.
+fn image_refs(body: &str) -> Vec<&str> {
+    let bytes = body.as_bytes();
+    let mut out = Vec::new();
+    let mut i = 0;
+    while i + 1 < bytes.len() {
+        // Look for the `![` opener.
+        if bytes[i] == b'!' && bytes[i + 1] == b'[' {
+            // Find the matching `]`.
+            if let Some(bracket_close) = body[i + 2..].find(']') {
+                let after_bracket = i + 2 + bracket_close + 1;
+                if after_bracket < bytes.len() && bytes[after_bracket] == b'(' {
+                    // Find the closing `)` of the link target.
+                    if let Some(paren_close) = body[after_bracket + 1..].find(')') {
+                        let target_start = after_bracket + 1;
+                        let target_end = target_start + paren_close;
+                        let target = &body[target_start..target_end];
+                        // Trim a possible ` "title"` suffix: take up to first space.
+                        let target = target.split_whitespace().next().unwrap_or("");
+                        out.push(target);
+                        i = target_end + 1;
+                        continue;
+                    }
+                }
+            }
+        }
+        i += 1;
+    }
+    out
+}
+
+/// Whether the named target is an **assemble-markdown** target (its steps
+/// contain at least one `Step::AssembleMarkdown` and no `Step::TypstCompile`) —
+/// i.e. a markdown deliverable that [`run_markdown_assemble_target`] builds
+/// rather than [`build`] compiling to PDF.
+///
+/// Loads the lesson read-only (ignoring diagnostics); an unloadable lesson or an
+/// unknown target returns `false`, so the caller falls through to the normal
+/// (typst) build path, which then reports the real error.
+pub fn target_is_markdown_assemble(root: &Path, target: &str) -> bool {
+    let (Some(lesson), _) = cph_model::load(root) else {
+        return false;
+    };
+    lesson
+        .targets
+        .iter()
+        .find(|t| t.name == target)
+        .is_some_and(|t| {
+            t.steps
+                .iter()
+                .any(|s| matches!(s, cph_model::Step::AssembleMarkdown { .. }))
+                && !t.steps
+                    .iter()
+                    .any(|s| matches!(s, cph_model::Step::TypstCompile { .. }))
+        })
+}
+
 /// Whether a part's folder exists on disk. cph-model emits `PartPathMissing`
 /// for parts whose folder is absent, so the orchestrator skips those parts in
 /// the structural/schema/coverage phases to avoid piling diagnostics on the
diff --git a/crates/cph-check/tests/pipeline.rs b/crates/cph-check/tests/pipeline.rs
index 12e51bc..a834da2 100644
--- a/crates/cph-check/tests/pipeline.rs
+++ b/crates/cph-check/tests/pipeline.rs
@@ -374,6 +374,283 @@ run = "printf SHOULD_NOT_RUN > build/ran.txt"
     cleanup(&tmp);
 }
 
+// --- assemble-markdown target (ADR-0015) ---------------------------------------
+
+/// Write a two-segment lesson where each segment carries a `slides.md` markdown
+/// content file, plus a `slides` target that assembles them. The `which` slice
+/// controls which segments get a `slides.md` (so the "skip absent" path can be
+/// exercised). Returns nothing; the caller runs `run_markdown_assemble_target`.
+fn write_markdown_assemble_target_lesson(tmp: &PathBuf, slides: &[(&str, &str)]) {
+    let mut parts = String::new();
+    for (i, (name, _)) in slides.iter().enumerate() {
+        if i > 0 {
+            parts.push('\n');
+        }
+        parts.push_str(&format!(
+            "[[parts]]\nkind = \"segment\"\npath = \"segments/{name}\"\n"
+        ));
+    }
+    std::fs::write(
+        tmp.join("manifest.toml"),
+        format!(
+            r#"
+[project]
+id = "md"
+name = "md"
+
+[info]
+title = "md"
+
+{parts}
+
+[targets.slides]
+artifact = {{ type = "single-file", filepath = "build/slides.md" }}
+[[targets.slides.steps]]
+type = "assemble-markdown"
+field = "slides"
+"#
+        ),
+    )
+    .unwrap();
+    for (name, body) in slides {
+        let dir = tmp.join("segments").join(name);
+        std::fs::create_dir_all(&dir).unwrap();
+        std::fs::write(dir.join("element.toml"), "kind = \"segment\"\n").unwrap();
+        std::fs::write(dir.join("textbook.typ"), "text.\n").unwrap();
+        std::fs::write(dir.join("slides.md"), body).unwrap();
+    }
+}
+
+#[test]
+fn target_is_markdown_assemble_detects_shape() {
+    let tmp = tempdir();
+    write_markdown_assemble_target_lesson(
+        &tmp,
+        &[("a", "# A"), ("b", "# B")],
+    );
+    assert!(cph_check::target_is_markdown_assemble(&tmp, "slides"));
+    // The student target doesn't exist here → falls through to typst path → false.
+    assert!(!cph_check::target_is_markdown_assemble(&tmp, "student"));
+    cleanup(&tmp);
+}
+
+#[test]
+fn run_markdown_assemble_target_concatenates_in_parts_order() {
+    let tmp = tempdir();
+    write_markdown_assemble_target_lesson(
+        &tmp,
+        &[("intro", "# 规则一\n- 范围"), ("rule2", "# 规则二\n$8\\times$")],
+    );
+
+    let report = cph_check::run_markdown_assemble_target(&tmp, &engine(), "slides");
+    assert!(report.ok, "assembly should succeed: {:#?}", report);
+    assert_eq!(report.outcomes.len(), 1);
+    let o = &report.outcomes[0];
+    assert!(o.ok());
+    assert_eq!(o.parts_read, 2);
+    assert_eq!(o.field, "slides");
+
+    let out = tmp.join("build").join("slides.md");
+    let body = std::fs::read_to_string(&out).unwrap();
+    // Both parts present, in declared order, joined by a blank line.
+    let intro_idx = body.find("规则一").unwrap();
+    let rule2_idx = body.find("规则二").unwrap();
+    assert!(intro_idx < rule2_idx, "parts must keep manifest order");
+    assert!(body.contains("$8\\times$"), "KaTeX source survives as text");
+    cleanup(&tmp);
+}
+
+#[test]
+fn run_markdown_assemble_target_skips_parts_without_the_field() {
+    // Only the first segment has a slides.md; the second is skipped (optional).
+    let tmp = tempdir();
+    let mut parts = String::new();
+    parts.push_str("[[parts]]\nkind = \"segment\"\npath = \"segments/a\"\n\n");
+    parts.push_str("[[parts]]\nkind = \"segment\"\npath = \"segments/b\"\n");
+    std::fs::write(
+        tmp.join("manifest.toml"),
+        format!(
+            r#"
+[project]
+id = "md"
+name = "md"
+
+[info]
+title = "md"
+
+{parts}
+
+[targets.slides]
+artifact = {{ type = "single-file", filepath = "build/slides.md" }}
+[[targets.slides.steps]]
+type = "assemble-markdown"
+field = "slides"
+"#
+        ),
+    )
+    .unwrap();
+    for name in ["a", "b"] {
+        let dir = tmp.join("segments").join(name);
+        std::fs::create_dir_all(&dir).unwrap();
+        std::fs::write(dir.join("element.toml"), "kind = \"segment\"\n").unwrap();
+        std::fs::write(dir.join("textbook.typ"), "text.\n").unwrap();
+    }
+    // Only `a` gets a slides.md.
+    std::fs::write(tmp.join("segments/a/slides.md"), "# A only\n").unwrap();
+
+    let report = cph_check::run_markdown_assemble_target(&tmp, &engine(), "slides");
+    assert!(report.ok, "{:#?}", report);
+    assert_eq!(report.outcomes[0].parts_read, 1);
+    let body = std::fs::read_to_string(tmp.join("build/slides.md")).unwrap();
+    // The assembler prepends the course title (info.title = "md") as an h1,
+    // then the single part's slides.md.
+    assert_eq!(body, "# md\n\n# A only\n");
+    cleanup(&tmp);
+}
+
+#[test]
+fn run_markdown_assemble_target_copies_referenced_images() {
+    // A slides.md that references two local images; both must be copied from
+    // the engineering root into the build root (same relative path) so the
+    // build dir is self-contained (ADR-0015).
+    let tmp = tempdir();
+    std::fs::write(
+        tmp.join("manifest.toml"),
+        r#"
+[project]
+id = "md"
+name = "md"
+
+[info]
+title = "md"
+
+[[parts]]
+kind = "segment"
+path = "segments/a"
+
+[targets.slides]
+artifact = { type = "single-file", filepath = "build/slides.md" }
+[[targets.slides.steps]]
+type = "assemble-markdown"
+field = "slides"
+"#,
+    )
+    .unwrap();
+    let dir = tmp.join("segments").join("a");
+    std::fs::create_dir_all(&dir).unwrap();
+    std::fs::write(dir.join("element.toml"), "kind = \"segment\"\n").unwrap();
+    std::fs::write(dir.join("textbook.typ"), "text.\n").unwrap();
+    std::fs::write(
+        dir.join("slides.md"),
+        "## A\n\n![one](assets/img/one.png)\n\n![two](assets/img/two.png)\n",
+    )
+    .unwrap();
+    // The two referenced images exist at the engineering root.
+    std::fs::create_dir_all(tmp.join("assets/img")).unwrap();
+    std::fs::write(tmp.join("assets/img/one.png"), b"PNG1").unwrap();
+    std::fs::write(tmp.join("assets/img/two.png"), b"PNG2").unwrap();
+
+    let report = cph_check::run_markdown_assemble_target(&tmp, &engine(), "slides");
+    assert!(report.ok, "{:#?}", report);
+
+    // Both images copied into the build root, preserving the assets/img/ path.
+    assert_eq!(
+        std::fs::read(tmp.join("build/assets/img/one.png")).unwrap(),
+        b"PNG1"
+    );
+    assert_eq!(
+        std::fs::read(tmp.join("build/assets/img/two.png")).unwrap(),
+        b"PNG2"
+    );
+    cleanup(&tmp);
+}
+
+#[test]
+fn run_markdown_assemble_target_fails_on_missing_referenced_image() {
+    // A slides.md references an image that does NOT exist at the engineering
+    // root → a build-process error (self-contained build would be broken),
+    // not a lesson diagnostic.
+    let tmp = tempdir();
+    std::fs::write(
+        tmp.join("manifest.toml"),
+        r#"
+[project]
+id = "md"
+name = "md"
+
+[info]
+title = "md"
+
+[[parts]]
+kind = "segment"
+path = "segments/a"
+
+[targets.slides]
+artifact = { type = "single-file", filepath = "build/slides.md" }
+[[targets.slides.steps]]
+type = "assemble-markdown"
+field = "slides"
+"#,
+    )
+    .unwrap();
+    let dir = tmp.join("segments").join("a");
+    std::fs::create_dir_all(&dir).unwrap();
+    std::fs::write(dir.join("element.toml"), "kind = \"segment\"\n").unwrap();
+    std::fs::write(dir.join("textbook.typ"), "text.\n").unwrap();
+    std::fs::write(
+        dir.join("slides.md"),
+        "## A\n\n![ghost](assets/img/ghost.png)\n",
+    )
+    .unwrap();
+    // No image created → reference is dangling.
+
+    let report = cph_check::run_markdown_assemble_target(&tmp, &engine(), "slides");
+    assert!(!report.ok, "a missing referenced image must fail the build");
+    let o = &report.outcomes[0];
+    assert!(o.error.as_deref().unwrap_or("").contains("ghost.png"));
+    // The markdown was still written, but the image was not (it doesn't exist).
+    assert!(tmp.join("build/slides.md").is_file());
+    cleanup(&tmp);
+}
+
+#[test]
+fn run_markdown_assemble_target_refuses_broken_lesson() {
+    // A part path that doesn't exist → check error → assembly never runs.
+    let tmp = tempdir();
+    std::fs::write(
+        tmp.join("manifest.toml"),
+        r#"
+[project]
+id = "md"
+name = "md"
+
+[info]
+title = "md"
+
+[[parts]]
+kind = "segment"
+path = "segments/missing"
+
+[targets.slides]
+artifact = { type = "single-file", filepath = "build/slides.md" }
+[[targets.slides.steps]]
+type = "assemble-markdown"
+field = "slides"
+"#,
+    )
+    .unwrap();
+
+    let report = cph_check::run_markdown_assemble_target(&tmp, &engine(), "slides");
+    assert!(!report.ok);
+    assert!(report.outcomes.is_empty(), "no assembly on a broken lesson");
+    assert!(report.check.has_errors());
+    assert!(
+        !tmp.join("build").join("slides.md").exists(),
+        "nothing should have been written"
+    );
+    cleanup(&tmp);
+}
+
 // --- tiny temp-dir helpers (no external dev-dep) ------------------------------
 
 fn tempdir() -> PathBuf {
diff --git a/crates/cph-cli/src/main.rs b/crates/cph-cli/src/main.rs
index 730427f..10a0936 100644
--- a/crates/cph-cli/src/main.rs
+++ b/crates/cph-cli/src/main.rs
@@ -110,6 +110,13 @@ fn run_build(
         return run_shell_build(path, engine, target);
     }
 
+    // A target whose steps assemble markdown (ADR-0015: slides outline / 逐字稿
+    // transcript surfaces) is built by concatenating per-element `<field>.md`
+    // files in parts order, not by compiling a typst template.
+    if cph_check::target_is_markdown_assemble(path, target) {
+        return run_markdown_assemble_build(path, engine, target);
+    }
+
     let out_path = out.unwrap_or_else(|| path.join("build").join(format!("{target}.pdf")));
 
     let (pdf, report) = cph_check::build(path, engine, target);
@@ -182,3 +189,43 @@ fn run_shell_build(path: &std::path::Path, engine: &Engine, target: &str) -> Exi
         ExitCode::FAILURE
     }
 }
+
+/// Run an assemble-markdown target: concatenate each element's `<field>.md`
+/// (in parts order) and write the single-file artifact (ADR-0015). Like the
+/// shell path, this is opt-in — only on an explicit `cph build --target <name>`,
+/// and never in `check`.
+fn run_markdown_assemble_build(path: &std::path::Path, engine: &Engine, target: &str) -> ExitCode {
+    eprintln!("assembling markdown target '{target}' (reading parts under {})", path.display());
+    let report = cph_check::run_markdown_assemble_target(path, engine, target);
+    print_diagnostics(&report.check);
+
+    if report.outcomes.is_empty() && !report.ok {
+        if report.check.has_errors() {
+            eprintln!("build refused: {} errors (fix the lesson first)", report.check.error_count());
+        } else {
+            eprintln!("build failed: target '{target}' has no assemble-markdown steps to run");
+        }
+        return ExitCode::FAILURE;
+    }
+
+    for outcome in &report.outcomes {
+        eprintln!(
+            "assembled field '{}' from {} part(s)",
+            outcome.field, outcome.parts_read
+        );
+        if let Some(out_rel) = &outcome.written {
+            println!("wrote {} ({} bytes)", path.join(out_rel).display(), outcome.body.len());
+        }
+        if let Some(err) = &outcome.error {
+            eprintln!("assemble failed: {err}");
+        }
+    }
+
+    if report.ok {
+        println!("assemble-markdown target '{target}' completed: {} step(s) ok", report.outcomes.len());
+        ExitCode::SUCCESS
+    } else {
+        eprintln!("assemble-markdown target '{target}' failed");
+        ExitCode::FAILURE
+    }
+}
diff --git a/crates/cph-model/src/lib.rs b/crates/cph-model/src/lib.rs
index 7b89748..b3c4d10 100644
--- a/crates/cph-model/src/lib.rs
+++ b/crates/cph-model/src/lib.rs
@@ -163,6 +163,7 @@ impl Artifact {
 /// inductive Step where
 ///   | typstCompile (template : String)
 ///   | shell (run : String)
+///   | assembleMarkdown (field : String)
 /// ```
 ///
 /// A step is a *typed* operation (extensible): `TypstCompile` compiles a
@@ -170,9 +171,13 @@ impl Artifact {
 /// the manifest into it (`--input manifest=…`), which a bare `typst compile`
 /// string cannot express. `Shell` is the escape hatch for steps that resist
 /// declaration (ADR-0005's medium-only category (b): HTML interactive / npm
-/// builds). The on-disk `[[steps]]` entry's `type` discriminator selects the
-/// variant: `"typst-compile"` → [`Step::TypstCompile`], `"shell"` →
-/// [`Step::Shell`].
+/// builds). `AssembleMarkdown` concatenates a per-element markdown content
+/// field into a single-file markdown deliverable (ADR-0015: the slides outline
+/// and 逐字稿 transcript surfaces, authored directly in markdown+KaTeX to
+/// sidestep the typst→md conversion). The on-disk `[[steps]]` entry's `type`
+/// discriminator selects the variant: `"typst-compile"` →
+/// [`Step::TypstCompile`], `"shell"` → [`Step::Shell`], `"assemble-markdown"`
+/// → [`Step::AssembleMarkdown`].
 ///
 /// As with [`Artifact`], this alignment is maintained by review, not CI.
 #[derive(Debug, Clone, PartialEq, Eq, Serialize)]
@@ -189,6 +194,15 @@ pub enum Step {
         /// The command line to run.
         run: String,
     },
+    /// Assemble a single-file markdown deliverable by concatenating each
+    /// element's `field` markdown content file in `[[parts]]` order. Mirrors
+    /// Lean `Step.assembleMarkdown` (ADR-0015). Not a typst build — the
+    /// framework owns the read/concatenate/write itself.
+    AssembleMarkdown {
+        /// The per-element markdown content field to assemble (e.g. `slides`,
+        /// `transcript`).
+        field: String,
+    },
 }
 
 impl Step {
@@ -901,7 +915,9 @@ fn parse_step(name: &str, value: toml::Value, diags: &mut Vec<Diagnostic>) -> Op
                     DiagCode::SchemaViolation,
                     format!("target '{name}' has a step with a non-string `type`"),
                 )
-                .with_hint("set the step `type` to \"typst-compile\" or \"shell\""),
+                .with_hint(
+                    "set the step `type` to \"typst-compile\", \"shell\", or \"assemble-markdown\"",
+                ),
             );
             return None;
         }
@@ -911,7 +927,9 @@ fn parse_step(name: &str, value: toml::Value, diags: &mut Vec<Diagnostic>) -> Op
                     DiagCode::SchemaViolation,
                     format!("target '{name}' has a step with no `type` discriminator"),
                 )
-                .with_hint("add `type = \"typst-compile\"` (or `\"shell\"`) to the step"),
+                .with_hint(
+                    "add `type = \"typst-compile\"` (or `\"shell\"`, `\"assemble-markdown\"`) to the step",
+                ),
             );
             return None;
         }
@@ -948,16 +966,33 @@ fn parse_step(name: &str, value: toml::Value, diags: &mut Vec<Diagnostic>) -> Op
                 None
             }
         },
+        "assemble-markdown" => match take_string_field(&mut table, "field") {
+            Some(field) => Some(Step::AssembleMarkdown { field }),
+            None => {
+                diags.push(
+                    Diagnostic::error(
+                        DiagCode::SchemaViolation,
+                        format!(
+                            "target '{name}' assemble-markdown step is missing a string `field`"
+                        ),
+                    )
+                    .with_hint("add `field = \"slides\"` (or `\"transcript\"`) to the step"),
+                );
+                None
+            }
+        },
         other => {
             diags.push(
                 Diagnostic::error(
                     DiagCode::SchemaViolation,
                     format!(
                         "target '{name}' has a step with an invalid `type` '{other}'; \
-                         valid values are \"typst-compile\", \"shell\""
+                         valid values are \"typst-compile\", \"shell\", \"assemble-markdown\""
                     ),
                 )
-                .with_hint("set the step `type` to \"typst-compile\" or \"shell\""),
+                .with_hint(
+                    "set the step `type` to \"typst-compile\", \"shell\", or \"assemble-markdown\"",
+                ),
             );
             None
         }
diff --git a/crates/cph-schema/schemas/example.json b/crates/cph-schema/schemas/example.json
index 01103d7..ee10cb0 100644
--- a/crates/cph-schema/schemas/example.json
+++ b/crates/cph-schema/schemas/example.json
@@ -6,6 +6,8 @@
   "properties": {
     "problem": { "type": "string", "x-cph-content": true },
     "solution": { "type": "string", "x-cph-content": true },
+    "slides": { "type": "string", "x-cph-content": "md" },
+    "transcript": { "type": "string", "x-cph-content": "md" },
     "source": { "type": "string" }
   },
   "required": ["problem", "solution"],
diff --git a/crates/cph-schema/schemas/lemma.json b/crates/cph-schema/schemas/lemma.json
index e21433f..e5cbd6c 100644
--- a/crates/cph-schema/schemas/lemma.json
+++ b/crates/cph-schema/schemas/lemma.json
@@ -5,7 +5,9 @@
   "type": "object",
   "properties": {
     "stmt": { "type": "string", "x-cph-content": true },
-    "proof": { "type": "string", "x-cph-content": true }
+    "proof": { "type": "string", "x-cph-content": true },
+    "slides": { "type": "string", "x-cph-content": "md" },
+    "transcript": { "type": "string", "x-cph-content": "md" }
   },
   "required": ["stmt"],
   "additionalProperties": false
diff --git a/crates/cph-schema/schemas/segment.json b/crates/cph-schema/schemas/segment.json
index 91f8a37..ea5d53f 100644
--- a/crates/cph-schema/schemas/segment.json
+++ b/crates/cph-schema/schemas/segment.json
@@ -4,7 +4,9 @@
   "title": "segment",
   "type": "object",
   "properties": {
-    "textbook": { "type": "string", "x-cph-content": true }
+    "textbook": { "type": "string", "x-cph-content": true },
+    "slides": { "type": "string", "x-cph-content": "md" },
+    "transcript": { "type": "string", "x-cph-content": "md" }
   },
   "required": ["textbook"],
   "additionalProperties": false
diff --git a/crates/cph-schema/schemas/sop.json b/crates/cph-schema/schemas/sop.json
index db6d944..8da118a 100644
--- a/crates/cph-schema/schemas/sop.json
+++ b/crates/cph-schema/schemas/sop.json
@@ -4,7 +4,9 @@
   "title": "sop",
   "type": "object",
   "properties": {
-    "sop": { "type": "string", "x-cph-content": true }
+    "sop": { "type": "string", "x-cph-content": true },
+    "slides": { "type": "string", "x-cph-content": "md" },
+    "transcript": { "type": "string", "x-cph-content": "md" }
   },
   "required": ["sop"],
   "additionalProperties": false
diff --git a/crates/cph-schema/src/lib.rs b/crates/cph-schema/src/lib.rs
index 968db7b..1932b49 100644
--- a/crates/cph-schema/src/lib.rs
+++ b/crates/cph-schema/src/lib.rs
@@ -116,12 +116,16 @@ struct ScalarField {
     required: bool,
 }
 
-/// One content property of a kind (a field whose value is a sibling `.typ`
-/// file, marked with `x-cph-content`).
+/// One content property of a kind (a field whose value is a sibling content
+/// file, marked with `x-cph-content`). The file extension is determined by the
+/// marker's value: `true` (or `"typ"`) → `.typ`; any other string → that
+/// extension (e.g. `"md"` → `.md`, ADR-0015 markdown content surfaces).
 #[derive(Debug, Clone)]
 struct ContentField {
     name: String,
     required: bool,
+    /// The content file's extension without the dot (e.g. `"typ"`, `"md"`).
+    ext: String,
 }
 
 /// A parsed kind schema: the content fields (which drive `.typ` existence
@@ -190,15 +194,22 @@ impl KindSchema {
             .unwrap_or_else(|| panic!("kind schema for '{kind}' has no `properties` object"));
 
         for (name, prop) in props {
-            let is_content = prop
-                .get(CONTENT_MARKER)
-                .and_then(Json::as_bool)
-                .unwrap_or(false);
+            // The `x-cph-content` marker selects a content field and fixes its
+            // file extension. `true` (the legacy form) and the string `"typ"`
+            // both mean a `.typ` file; any other string is taken as the
+            // extension (e.g. `"md"` → `.md` for markdown content surfaces,
+            // ADR-0015). Absent ⇒ a scalar field.
+            let ext = match prop.get(CONTENT_MARKER) {
+                Some(Json::Bool(true)) => Some("typ".to_string()),
+                Some(Json::String(s)) => Some(s.clone()),
+                _ => None,
+            };
 
-            if is_content {
+            if let Some(ext) = ext {
                 content_fields.push(ContentField {
                     name: name.clone(),
                     required: is_required(name),
+                    ext,
                 });
             } else {
                 let ty = prop
@@ -307,14 +318,8 @@ fn validate_scalars(schema: &KindSchema, desc: &ElementDescriptor, diags: &mut V
     // Each present scalar: type / enum checks, and (if known) reject unexpected
     // keys when additionalProperties is false. Content fields must never appear
     // in element.toml.
-    let content_names: Vec<&str> = schema
-        .content_fields
-        .iter()
-        .map(|f| f.name.as_str())
-        .collect();
-
     for (key, value) in scalars {
-        if content_names.contains(&key.as_str()) {
+        if let Some(field) = schema.content_fields.iter().find(|f| f.name == *key) {
             diags.push(
                 Diagnostic::error(
                     DiagCode::SchemaViolation,
@@ -324,7 +329,8 @@ fn validate_scalars(schema: &KindSchema, desc: &ElementDescriptor, diags: &mut V
                     ),
                 )
                 .with_hint(format!(
-                    "remove '{key}' from element.toml; put its content in the sibling file '{key}.typ'"
+                    "remove '{key}' from element.toml; put its content in the sibling file '{key}.{}'",
+                    field.ext
                 )),
             );
             continue;
@@ -390,8 +396,9 @@ fn validate_scalars(schema: &KindSchema, desc: &ElementDescriptor, diags: &mut V
     }
 }
 
-/// For each required content field, check that its `<field>.typ` file exists in
-/// the element directory.
+/// For each required content field, check that its `<field>.<ext>` file exists
+/// in the element directory (`.typ` for typst content, `.md` for markdown
+/// content surfaces — ADR-0015).
 fn validate_content_files(
     schema: &KindSchema,
     desc: &ElementDescriptor,
@@ -402,7 +409,7 @@ fn validate_content_files(
             // Optional content (e.g. lemma's `proof`): absence is fine.
             continue;
         }
-        let file_name = format!("{}.typ", field.name);
+        let file_name = format!("{}.{}", field.name, field.ext);
         let path = desc.dir.join(&file_name);
         if !path.is_file() {
             diags.push(
@@ -415,7 +422,8 @@ fn validate_content_files(
                     ),
                 )
                 .with_hint(format!(
-                    "create '{file_name}' in the element folder with the typst source for '{}'",
+                    "create '{file_name}' in the element folder with the {} source for '{}'",
+                    ext_label(&field.ext),
                     field.name
                 )),
             );
@@ -423,6 +431,15 @@ fn validate_content_files(
     }
 }
 
+/// A human label for a content-file extension, used in diagnostic hints.
+fn ext_label(ext: &str) -> &'static str {
+    match ext {
+        "md" => "markdown",
+        "typ" => "typst",
+        _ => "content",
+    }
+}
+
 /// Convert a `toml::Value` scalar into the equivalent `serde_json::Value` for
 /// type checking. Composite TOML values (arrays / tables) are converted
 /// structurally too, though the MVP stdlib schemas only use scalars.
@@ -484,7 +501,14 @@ mod tests {
     #[test]
     fn example_field_split() {
         let s = schema_for("example").unwrap();
-        assert_eq!(s.content_field_names(), vec!["problem", "solution"]);
+        // `problem`/`solution` are typst content; `slides`/`transcript` are the
+        // optional markdown content surfaces (ADR-0015); `source` is a scalar.
+        // (serde_json's Map is a BTreeMap without the `preserve_order` feature,
+        // so content field names come back alphabetically sorted.)
+        assert_eq!(
+            s.content_field_names(),
+            vec!["problem", "slides", "solution", "transcript"]
+        );
         assert_eq!(s.scalar_field_names(), vec!["source"]);
     }
 
@@ -496,4 +520,29 @@ mod tests {
         let stmt = s.content_fields.iter().find(|f| f.name == "stmt").unwrap();
         assert!(stmt.required);
     }
+
+    #[test]
+    fn markdown_content_fields_carry_md_ext() {
+        // ADR-0015: `slides`/`transcript` are markdown content fields (`x-cph-content: "md"`),
+        // distinct from the `.typ` content fields.
+        let s = schema_for("segment").unwrap();
+        let slides = s.content_fields.iter().find(|f| f.name == "slides").unwrap();
+        assert_eq!(slides.ext, "md");
+        assert!(!slides.required);
+        let transcript = s
+            .content_fields
+            .iter()
+            .find(|f| f.name == "transcript")
+            .unwrap();
+        assert_eq!(transcript.ext, "md");
+        assert!(!transcript.required);
+        // Legacy typst content still resolves to `.typ`.
+        let textbook = s
+            .content_fields
+            .iter()
+            .find(|f| f.name == "textbook")
+            .unwrap();
+        assert_eq!(textbook.ext, "typ");
+        assert!(textbook.required);
+    }
 }
diff --git a/crates/cph-typst/src/lib.rs b/crates/cph-typst/src/lib.rs
index a904693..e1bf25f 100644
--- a/crates/cph-typst/src/lib.rs
+++ b/crates/cph-typst/src/lib.rs
@@ -220,11 +220,13 @@ fn target_precheck(lesson: &Lesson, target: &str) -> Result<PathBuf, Vec<Diagnos
 }
 
 /// The template of the first [`Step::TypstCompile`] in `tc`, if any (MVP: a
-/// target has one step; a `Shell`-only target yields `None`).
+/// target has one step; a `Shell`-only or `AssembleMarkdown`-only target yields
+/// `None`).
 fn template_step(tc: &TargetConfig) -> Option<PathBuf> {
     tc.steps.iter().find_map(|s| match s {
         Step::TypstCompile { template } => Some(template.clone()),
         Step::Shell { .. } => None,
+        Step::AssembleMarkdown { .. } => None,
     })
 }
 
diff --git a/docs/adr/0015-slides-and-transcript-markdown-surfaces.md b/docs/adr/0015-slides-and-transcript-markdown-surfaces.md
new file mode 100644
index 0000000..14fb1e0
--- /dev/null
+++ b/docs/adr/0015-slides-and-transcript-markdown-surfaces.md
@@ -0,0 +1,177 @@
+# ADR 0015: Slides & Transcript As Per-Element Markdown Content Surfaces
+
+## Status
+
+Accepted — and to be implemented. Adds two new **markdown** content surfaces
+(`slides`, `transcript`) as per-element optional content fields, plus a typed
+`Step.assembleMarkdown` that concatenates them in parts order into a single-file
+markdown deliverable.
+
+Builds on **ADR-0005** ((a)-class fields like 逐字稿 live inside a kind's
+`ElementData`; (b)-class medium lives in the target build), **ADR-0006**
+(`content` leaves in a kind schema), **ADR-0011** (a target is an artifact +
+ordered typed steps; presentation lives in the template/content not the
+manifest), **ADR-0013** (non-typst targets ride a build step; failures are
+build-process errors outside the 6-class taxonomy), and **ADR-0014** (the
+markdown/HTML export direction — R1 shared typst-content vs. R2 target-specific
+content; math must survive as markup not images).
+
+## Context
+
+The KenKen lesson (and curriculum generally) needs a **slides** deliverable —
+an **outline-grade** projection (titles + bullet points; detail belongs in the
+讲义/教案 or the 逐字稿), aimed at a third-party platform whose interchange
+format is a **specific markdown dialect**. It also needs a **逐字稿**
+(transcript) — the spoken text, read aloud, no typography needed.
+
+Two prior facts made the design:
+
+1. **slides content ≠ 讲义 content.** ADR-0014 already recorded that a
+   markdown/HTML target's content does not fully overlap the PDF 讲义/教案.
+   Forcing slides to be a projection of the same typst source (ADR-0014's R1)
+   drags in the hard **typst→markdown math conversion** (typlite renders math
+   as images; 0.15 MathML is accessibility-oriented and hard to turn into KaTeX;
+   `mitex` is LaTeX→typst, the reverse direction — ADR-0014 §Context).
+2. **slides/逐字稿 can be authored directly in markdown.** If `slides` and
+   `transcript` are authored as **markdown + KaTeX from the start**, there is
+   *no* typst→markdown conversion at all — the math is KaTeX source (`$…$`) on
+   the day it is written. The math principle ("markup, not images", ADR-0014)
+   is satisfied for free, not by building a fragile converter.
+
+So for the slides/transcript surfaces this ADR picks **ADR-0014's R2**
+(target-specific content) deliberately: it **sidesteps** the typst→md converter
+entirely. The converter (R1) remains an open problem only for the *讲义* surface,
+which is independent and out of scope here.
+
+The spec already pins where (a)-class fields live: `Element.lean` says
+"逐字稿、重点圈划 落在具体 kind 的 `ElementData` 内". So `slides`/`transcript`
+are modeled as **per-kind optional `content` leaves** (declared in each kind's
+JSON schema), consistent with the existing `textbook`/`problem`/`solution`
+content fields — except their **format is markdown, not typst**. This is a real
+extension to the `content` concept: a content leaf now carries a *format*
+(typst | markdown). `RichContent.lean` gains a `ContentFormat` and the content
+ref gains a `format` field.
+
+## Decision
+
+### Two new markdown content fields on every kind
+
+Each stdlib kind schema (`segment`, `example`, `lemma`, `sop`) gains two
+**optional** `content` fields — `slides` and `transcript` — whose format is
+**markdown** (`x-cph-content: "md"`, the new extension-bearing form of the
+marker; the old `x-cph-content: true` stays `.typ` for backward compat).
+Optional ⇒ a kind instance without them is legal (like a lemma's `proof`).
+Convention: `slides.md` is **outline-grade** (title + bullets; no worked
+detail — that lives in 讲义/逐字稿); `transcript.md` is the **spoken** text.
+
+### A new typed step `Step.assembleMarkdown field`
+
+A target like `[targets.slides]` carries `[[…steps]] type = "assemble-markdown"
+field = "slides"` over a `single-file` artifact. The framework **owns** the
+assembly (not a `cat`): it walks the manifest `[[parts]]` **in order**, reads
+each element's `<field>.md` if present (skips if absent), joins them with a
+blank-line separator, and writes the single file to the artifact's `filepath`
+(relative to the engineering root). It does **not** inject headings — each
+`slides.md`/`transcript.md` authors its own `# …` (presentation lives in the
+content, ADR-0011).
+
+This is **typed**, not `shell`, because ordered-read + skip-missing +
+write-to-artifact-path is cleaner owned by the framework than expressed as a
+shell pipeline, and it stays deterministic/testable.
+
+### Self-contained build dir: the assembler collects referenced images
+
+The assembled `slides.md` is written under the build root (the artifact
+`filepath`'s parent, e.g. `build/`). For that directory to be a **standalone
+deliverable** — uploadable to the platform on its own — its `![](rel)` image
+references must resolve *within* it, not back into the engineering root. So the
+assembler, after writing the markdown, scans it for local `![](rel)` references
+and copies each referenced image from the engineering root (`<eng-root>/<rel>`)
+into the build root (`<build-root>/<rel>`), preserving the relative path. This
+is part of the `assembleMarkdown` step's owned responsibility (one step, not a
+separate copy step), keeping a slides target a single typed step.
+
+Conventions that fall out of this:
+
+- **Image references are authored relative to the engineering root** (`assets/images/…`),
+  which is *also* their form relative to the build root after the copy — so the
+  same path string is correct both at authoring time (the source image exists at
+  `<eng-root>/assets/images/…`) and in the assembled output.
+- **Widget references are build-relative** (`interactives/ex1.html`): widgets are
+  *build artifacts* produced by a separate target (the `interactives` shell
+  target writes `build/interactives/`), not engineering-root sources, so they are
+  referenced but **not** copied by the assembler. A fully self-contained `build/`
+  therefore requires running the `interactives` target first — a documented
+  prerequisite, like `kendoku` reachability (ADR-0013).
+- **External URLs** (`http://`/`https://`/`data:`) are left untouched.
+- A referenced image **missing** at the engineering root is a build-process error
+  (the build dir would be broken), reported by the build layer — it does **not**
+  enter the 6-class taxonomy.
+
+### Course-level h1 is injected
+
+The assembler prepends the course title (from manifest `[info] title`) as the
+document's `#` h1. This is course metadata, not per-element content; each
+element's `<field>.md` then contributes its `##` part and `###` 小节. This is
+the one piece of presentation the assembler owns at the document level (ADR-0011's
+"presentation lives in the content" governs *per-element* presentation).
+
+### Same three boundaries as the shell step (ADR-0013)
+
+`assembleMarkdown` is a **non-typst target**, identical in boundary to the
+shell step:
+
+1. **opt-in by construction** — runs only on an explicit
+   `cph build --target <name>`, never in `check`. `check` validates lesson
+   *structure*, not the outcome of assembling markdown.
+2. **failure is a build-process error, not a lesson diagnostic** — an assembly
+   failure (e.g. write error) does **not** enter the 6-class diagnostic
+   taxonomy; it is reported by the build layer. We do **not** add an
+   `AssembleMarkdown` diagnostic code (taxonomy stays 6, ADR-0010/0012/0013).
+3. **non-typst targets skip typst compile** — `check`'s compile phase already
+   filters to `TypstCompile`-bearing targets, so a pure `assembleMarkdown`
+   target is naturally excluded.
+
+### Single-file artifact now; structured FileTree deferred
+
+The artifact is a **single-file** markdown. The user reasoned that until parts
+are organized as an **ordered tree** (rather than the current flat ordered
+list), single-file is unavoidable — a loose FileTree of per-element `.md`
+files would lose the ordering/structure the manifest encodes. So: single-file
+now; a structured multi-file FileTree artifact is deferred until the parts
+tree-restructure lands (it is out of scope here and recorded as OPEN).
+
+## Consequences
+
+- The framework gains two orthogonal content surfaces (outline slides, spoken
+  transcript) authored directly in markdown+KaTeX, with **no** typst→md math
+  conversion in their path. The math problem (ADR-0014) is *not* solved
+  generally — it is *avoided* for these surfaces by choosing R2.
+- A third typed `Step` variant (`assembleMarkdown`) joins `typstCompile` and
+  `shell`; the step universe is extensible by design (ADR-0011).
+- The `content` leaf concept broadens: it now carries a *format*. The
+  `x-cph-content` marker accepts an extension (`"md"`), backward-compatible
+  with the boolean form (`.typ`). `RichContent.lean` pins `ContentFormat`.
+- The 6-class diagnostic taxonomy is unchanged.
+- The 恒一小奥 KenKen lesson gains `slides`/`transcript` targets alongside its
+  existing `student`/`teacher`/`interactives` targets — all from one
+  engineering file through one CLI.
+
+## Open Questions / Deferred
+
+- **Third-party markdown dialect.** Baseline CommonMark + KaTeX (`$…$`/`$$…$$`)
+  ships now; the platform's specific dialect (page-break markers, callout /
+  fenced-div conventions, heading-depth rules, math-delimiter quirks) is a
+  transform layer to be added once the lesson's output is fed to the platform
+  and the real requirements are observed.
+- **Structured FileTree artifact.** A multi-file markdown deliverable organized
+  by the parts tree — deferred until the parts tree-restructure (itself a
+  separate OPEN / future ADR; touches the manifest schema, render assembly,
+  and all examples).
+- **讲义 (typst) → markdown.** Still ADR-0014's open R1/converter problem,
+  *unaffected* by this ADR. Whether the 讲义 surface ever needs a markdown
+  export is an independent decision.
+- **Universal vs. per-kind markdown surfaces.** `slides`/`transcript` are
+  declared per-kind (4 schemas, duplicated) to match ADR-0005's "(a)-class
+  lives in kind `ElementData`". Whether they should become a cross-cutting
+  universal content surface is reconsidered alongside the parts tree-restructure.
diff --git a/spec/Spec/Courseware.lean b/spec/Spec/Courseware.lean
index 6b5ade8..253a5b9 100644
--- a/spec/Spec/Courseware.lean
+++ b/spec/Spec/Courseware.lean
@@ -6,7 +6,7 @@ import Spec.Courseware.Open
 /-!
 # Courseware —— 产品层契约(课程工程文件)
 
-护城河:课程"工程文件"的语义母本与合法性规则。决策出处 ADR-0005..0014。按四组组织:
+护城河:课程"工程文件"的语义母本与合法性规则。决策出处 ADR-0005..0015。按四组组织:
 
 - **`Model`** —— 工程文件的内容模型:基元 `Primitives`、富内容 `RichContent`、原子
   单位 `Element`、单节课 `Lesson`(element 的有序序列)。
diff --git a/spec/Spec/Courseware/Export/Render.lean b/spec/Spec/Courseware/Export/Render.lean
index 866a6e0..ed71d5c 100644
--- a/spec/Spec/Courseware/Export/Render.lean
+++ b/spec/Spec/Courseware/Export/Render.lean
@@ -11,6 +11,9 @@ ADR-0009:export target 是一次 build,产出一个有类型的 `Artifact`。ADR
   *typed* 而非裸 shell,正因框架要把 **manifest 注入**模板(经 `--input manifest=…`),
   裸字符串表达不了这个 wiring。presentation(编号、样式)住模板里,不在 manifest。
 - `shell run` —— 逃生口,给难以声明的步骤(ADR-0005 的 (b) 类 medium-only)。
+- `assembleMarkdown field` —— 按 parts 顺序把每个 element 的 `field`(markdown content 叶子,
+  ADR-0015)拼成**单文件 markdown** 产物。typed 而非 `cat`,因为按 manifest `[[parts]]` 顺序读取 +
+  跳过缺失项 + 写到单文件产物路径这件事,框架 own 比裸 shell 稳。
 
 **渲染覆盖**:ADR-0011 废止了 per-target `RenderRule` 载荷——渲染的"how"已移进模板。
 契约只保留覆盖声明 `covers`(该 target 渲染哪些 kind),供种子诊断用。
@@ -25,6 +28,17 @@ ADR-0009:export target 是一次 build,产出一个有类型的 `Artifact`。ADR
    而由 build 执行层报告。诊断分类保持 6 类不变(ADR-0013 显式拒绝新增 `ShellStep` 诊断码)。
 3. **非-typst target 不过 typst 编译** —— 一个只含 `shell` step 的 target(教具包即此)由
    执行器跑命令,而非走 typst 引擎;`check` 的 compile 阶段跳过它。
+
+**assembleMarkdown step 的执行语义(ADR-0015)。** 与 `shell` 同属"非-typst target":框架 own
+读取每个 element 的 `<field>.md`、按 `[[parts]]` 顺序拼接、写到单文件产物路径(产物是 markdown,
+不经 typst 引擎)。同 shell 的三条边界:① 只在显式 `build --target` 跑,`check` 不跑;② 装配失败
+(如写盘失败、引用图缺失)是 build-过程错误,不入 6 类诊断;③ 非-typst target,`check` 的 compile
+阶段跳过它。**单文件产物**(现):装配器注入课程级标题为文档 h1(课程元数据,非 element 内容),
+每份 `slides.md`/`transcript.md` 贡献其 `##` part 与 `###` 小节(presentation 在内容里, ADR-0011),
+按 `[[parts]]` 顺序拼接(分隔符空行)。**自包含**:装配器扫描产物里的 `![](rel)` 图引用,把引用到的
+本地图从工程根按原相对路径复制进产物所在 build 根,使该 build 目录可独立交付;引用图在工程根缺失属
+build-过程错误(不入 6 类)。外部 URL(`http(s)://`/`data:`)不复制。结构化 FileTree 产物延后
+(待 parts 树形重组, ADR-0015 OPEN)。
 -/
 
 namespace Spec.Courseware
@@ -42,6 +56,13 @@ inductive Step where
       为 cwd 执行,opt-in(只在显式 build 该 target 时跑,`check` 不跑),失败属 build-过程错误
       而非 lesson 诊断。教具包(如 KenKen 交互 HTML 由外部 `kendoku` 生成)即走此 step。 -/
   | shell (run : String)
+  /-- 装配 markdown:按 `[[parts]]` 顺序读取每个 element 的 `field`(markdown content 叶子,
+      ADR-0015),注入课程 h1 后拼接成**单文件 markdown** 产物,并收集 `![](rel)` 引用的本地图进
+      build 根使产物自包含。typed 而非 `cat`:按序读取+跳过缺失+注入标题+写盘+收集图由框架 own。
+      **已实现**(ADR-0015):非-typst target(`check` 不跑,失败属 build-过程错误不入 6 类诊断)。
+      slides 大纲面 / 逐字稿口播面即走此 step(直接 markdown+KaTeX 撰写,绕开 typst→md 公式转换,
+      ADR-0014 R2)。 -/
+  | assembleMarkdown (field : String)
 
 /-- 一个 export target 的 build 规格(`PINNED` artifact + 有序 steps, ADR-0011)。 -/
 structure TargetSpec where
diff --git a/spec/Spec/Courseware/Model/RichContent.lean b/spec/Spec/Courseware/Model/RichContent.lean
index 05589bd..d1a8899 100644
--- a/spec/Spec/Courseware/Model/RichContent.lean
+++ b/spec/Spec/Courseware/Model/RichContent.lean
@@ -1,14 +1,20 @@
 /-!
 # RichContent —— 富内容(ADR-0006 的 prose 母本)
 
-ADR-0006:element schema 的"叶子"可以是 `content` 类型,其值是**一段 typst 源**,
-语义取该源作为 module 求值后的 body content。关键约束(均 ADR-0006,源自 typst 源码
-事实):富内容**不可无主**——typst 的源必须有 `FileId`,否则 span 脱锚、相对 import
-报"cannot access file system from here"。故每段富内容是 World 里的一等文件,坐落在
-一个**虚拟路径**上;相对 import 限本工程路径结构内 + `@package`(不跨工程)。
+ADR-0006:element schema 的"叶子"可以是 `content` 类型,其值是一段**源文本**,
+按其 **format** 决定语义(ADR-0015)。两种 format:
+- **typst** —— 一段 typst 源,语义取该源作为 module 求值后的 body content(讲义/教案面)。
+- **markdown** —— 一段**原样**的 markdown + KaTeX 源,**不经 typst 求值**(slides 大纲面 / 逐字稿口播面;
+  ADR-0015)。直接以 markdown 撰写**绕开** typst→markdown 的公式转换难题(ADR-0014 R2):公式一开始就是
+  KaTeX 源(`$…$`),没有"把 typst 公式转成 md"这一步。
 
-本模块只立 prose 锚点 + 最小抽象签名:typst 的 `Content`/`Module` 内部结构、JSON
-Schema 形状属实现细节,不进 Lean,只承诺"富内容由一个虚拟路径定位"这一条关系。
+关键约束(均 ADR-0006,源自 typst 源码事实):**typst** format 的富内容**不可无主**——typst 的源必须有
+`FileId`,否则 span 脱锚、相对 import 报"cannot access file system from here"。故每段 typst 富内容是 World 里
+的一等文件,坐落在一个**虚拟路径**上;相对 import 限本工程路径结构内 + `@package`(不跨工程)。markdown format
+的富内容不参与 typst 求值,但同样由一个虚拟路径定位(供 markdown 装配 step 按序读取,见 `Export/Render`)。
+
+本模块只立 prose 锚点 + 最小抽象签名:typst 的 `Content`/`Module` 内部结构、JSON Schema 形状、format 的
+具体判别属实现细节,不进 Lean,只承诺"富内容由一个虚拟路径定位"+"叶子带 format"这两条关系。
 -/
 
 namespace Spec.Courseware
@@ -18,11 +24,21 @@ namespace Spec.Courseware
 (ADR-0007),不在本层承诺,故 opaque。 -/
 opaque VPath : Type
 
-/-- 对一段富内容的**引用**:它坐落在某个虚拟路径上(`PINNED` 关系, ADR-0006)。
-刻意**不**建模源文本、不建模求值出的 `Content`(那是实现侧的事);只钉"富内容经由一个
-`VPath` 定位",作为 `Primitives.ElementData` 里 `content` 叶子的语义锚点。 -/
+/-- 富内容的 **format**(`PINNED`, ADR-0015)。`content` 叶子带 format:typst 叶子被 typst 求值;
+markdown 叶子原样保留(markdown+KaTeX 源,不经求值)。 -/
+inductive ContentFormat where
+  /-- typst 源:求值为 typst `Content`(讲义/教案面)。 -/
+  | typst
+  /-- markdown + KaTeX 源:原样保留,不经 typst 求值(slides 大纲面 / 逐字稿口播面;ADR-0015)。 -/
+  | markdown
+
+/-- 对一段富内容的**引用**:它坐落在某个虚拟路径上(`PINNED` 关系, ADR-0006),并带一个
+**format**(`PINNED`, ADR-0015)。刻意**不**建模源文本、不建模求值出的 `Content`(那是实现侧的事);
+只钉"富内容经由一个 `VPath` 定位 + 带 format",作为 `Primitives.ElementData` 里 `content` 叶子的语义锚点。 -/
 structure RichContentRef where
   /-- 该富内容所在的虚拟路径(ADR-0006;落盘后为真实相对路径, ADR-0007)。 -/
   vpath : VPath
+  /-- 该富内容的 format(ADR-0015)。 -/
+  format : ContentFormat
 
 end Spec.Courseware