diff --git a/internal/cli/job.go b/internal/cli/job.go
index 86ce40d..ccb0e3c 100644
--- a/internal/cli/job.go
+++ b/internal/cli/job.go
@@ -74,7 +74,7 @@ var jobRunCmd = &cobra.Command{
 			peers := engine.NewPeerRegistry()
 			dispatcher := engine.NewDispatcher(newLogger(), store.NewCapacityRepo(db), peers, exec)
 			specBytes, _ := json.Marshal(map[string]any{
-				"name":    spec.Job.Name,
+				"name":    spec.Name,
 				"command": "/bin/true", // placeholder; full HCL dispatch lands in a later phase
 			})
 			jobID, nodeID, err := dispatcher.Submit(ctx, runTarget, specBytes, runIDKey)
@@ -93,11 +93,11 @@ var jobRunCmd = &cobra.Command{
 
 		job := &model.Job{
 			ID:     uuid.NewString(),
-			Name:   spec.Job.Name,
+			Name:   spec.Name,
 			Spec:   args[0],
 			Status: model.JobStatusPending,
 		}
-		if err := exec.Run(ctx, job, toTaskSpecs(spec.Tasks)); err != nil {
+		if err := exec.Run(ctx, job, workloadToTaskSpecs(spec)); err != nil {
 			if jsonOutput {
 				_ = printJSON(map[string]any{"id": job.ID, "status": "failed", "error": err.Error()})
 				return err
@@ -330,3 +330,22 @@ func toTaskSpecs(in []jobspec.TaskSpec) []engine.TaskSpec {
 	}
 	return out
 }
+
+// workloadToTaskSpecs converts a *WorkloadSpec into the engine.TaskSpec
+// slice consumed by the executor. For the HCL adapter path the runtime
+// block carries the legacy task[0].Command; for the Markdown path the
+// runtime block is the canonical runtime abstraction (P07 will expand
+// this). When Runtime is nil we emit a single no-op task to preserve
+// the legacy "at least one task" invariant.
+func workloadToTaskSpecs(spec *jobspec.WorkloadSpec) []engine.TaskSpec {
+	if spec == nil {
+		return nil
+	}
+	if spec.Runtime == nil {
+		return []engine.TaskSpec{{Name: spec.Name, Command: "/bin/true"}}
+	}
+	return []engine.TaskSpec{{
+		Name:    spec.Name,
+		Command: spec.Runtime.Command,
+	}}
+}
diff --git a/internal/jobspec/dispatch.go b/internal/jobspec/dispatch.go
new file mode 100644
index 0000000..e025b9c
--- /dev/null
+++ b/internal/jobspec/dispatch.go
@@ -0,0 +1,147 @@
+package jobspec
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+// ParseFile reads a jobspec file from disk and dispatches on file
+// extension (R-013, REQ-064):
+//
+//   - .md  → ParseMarkdown (canonical Markdown+frontmatter, R-014/R-015)
+//   - .yaml/.yml → ParseMarkdown with the whole file treated as
+//     frontmatter and Body = "" (pure YAML, no Markdown body)
+//   - .hcl → ParseHCL (legacy adapter; wraps the existing HCL parser
+//     and converts Spec{Job, Tasks} into *WorkloadSpec with Kind="Job",
+//     REQ-090 migration window)
+//
+// Unknown extensions return an error. The dispatcher preserves
+// `orca job run old-spec.hcl` during the v0.9→v0.10 migration window
+// (REQ-090).
+func ParseFile(path string) (*WorkloadSpec, error) {
+	data, err := os.ReadFile(path)
+	if err != nil {
+		return nil, fmt.Errorf("read spec file: %w", err)
+	}
+	return Dispatch(data, filepath.Base(path))
+}
+
+// ParseHCLFile reads an HCL file and parses it via the legacy HCL parser,
+// returning the legacy *Spec. It is a convenience wrapper retained for
+// tests and direct HCL consumers that need the raw Spec{Job, Tasks}
+// shape during the v0.9→v0.10 migration window (REQ-090).
+//
+// Deprecated: use ParseFile (dispatcher) for new code. HCL is legacy per
+// R-013.
+func ParseHCLFile(path string) (*Spec, error) {
+	data, err := os.ReadFile(path)
+	if err != nil {
+		return nil, fmt.Errorf("read spec file: %w", err)
+	}
+	return ParseHCLLegacy(data, filepath.Base(path))
+}
+
+// Dispatch routes raw jobspec bytes on file extension to the
+// appropriate parser. filename is used only for HCL (the HCL decoder
+// needs a filename for error messages and syntax sniffing).
+func Dispatch(data []byte, filename string) (*WorkloadSpec, error) {
+	ext := strings.ToLower(filepath.Ext(filename))
+	switch ext {
+	case ".md":
+		return ParseMarkdown(data)
+	case ".yaml", ".yml":
+		// Pure YAML file: no Markdown body. Treat the whole file as
+		// the frontmatter block. Body is empty (R-015: no body to
+		// preserve).
+		spec, err := parseYAMLFile(data)
+		if err != nil {
+			return nil, err
+		}
+		return spec, nil
+	case ".hcl":
+		return ParseHCL(data, filename)
+	default:
+		return nil, fmt.Errorf("parse jobspec: unknown extension %q (want .md, .yaml, .yml, or .hcl)", ext)
+	}
+}
+
+// parseYAMLFile treats the whole file as a frontmatter block (no
+// surrounding `---` delimiters, no Markdown body). This routes .yaml
+// and .yml files through the same hand-rolled parser as .md.
+func parseYAMLFile(data []byte) (*WorkloadSpec, error) {
+	block := string(data)
+	if strings.TrimSpace(block) == "" {
+		return nil, fmt.Errorf("parse yaml: empty file")
+	}
+	spec, err := parseFrontmatterBlock(block)
+	if err != nil {
+		return nil, err
+	}
+	spec.Body = ""
+	if err := validateWorkload(spec); err != nil {
+		return nil, err
+	}
+	return spec, nil
+}
+
+// ParseHCL parses a legacy HCL jobspec and adapts it into a *WorkloadSpec
+// (REQ-064 adapter, REQ-090 migration window). The existing HCL
+// Spec{Job, Tasks} shape is converted to:
+//
+//	Kind:     "Job"
+//	Name:     spec.Job.Name
+//	Runtime:  {one_of: "process", command: tasks[0].Command}
+//
+// Body is empty (HCL has no Markdown body). The legacy Spec struct and
+// ParseHCLLegacy are retained for direct HCL consumers that have not yet
+// migrated.
+//
+// Deprecated: use the dispatcher (ParseFile/Dispatch). HCL is legacy
+// per R-013; the HCL path is retained only for the v0.9→v0.10 migration
+// window (REQ-090) and will be removed in v1.0.
+func ParseHCL(data []byte, filename string) (*WorkloadSpec, error) {
+	spec, err := ParseHCLLegacy(data, filename)
+	if err != nil {
+		return nil, err
+	}
+	ws := &WorkloadSpec{
+		SpecVersion: "",
+		Kind:        "Job",
+		Name:        spec.Job.Name,
+		Count:       1,
+		Body:        "",
+	}
+	if len(spec.Tasks) > 0 {
+		ws.Runtime = &RuntimeBlock{
+			OneOf:   "process",
+			Command: spec.Tasks[0].Command,
+		}
+	}
+	return ws, nil
+}
+
+// ParseHCLLegacy is the original HCL-only parser retained for direct
+// HCL consumers (e.g. the cli/job.go toTaskSpecs path during the
+// migration window). New code should call ParseHCL (which returns a
+// *WorkloadSpec) or the dispatcher. Deprecated: HCL is legacy per
+// R-013; see ParseHCL.
+func ParseHCLLegacy(data []byte, filename string) (*Spec, error) {
+	var spec Spec
+	if err := hclDecode(filename, data, &spec); err != nil {
+		return nil, fmt.Errorf("decode hcl: %w", err)
+	}
+	if spec.Job.Name == "" {
+		return nil, fmt.Errorf("spec missing job name")
+	}
+	if len(spec.Tasks) == 0 {
+		return nil, fmt.Errorf("spec must have at least one task")
+	}
+	for i, t := range spec.Tasks {
+		if t.Command == "" {
+			return nil, fmt.Errorf("task[%d] (%s) missing command", i, t.Name)
+		}
+	}
+	return &spec, nil
+}
diff --git a/internal/jobspec/dispatch_test.go b/internal/jobspec/dispatch_test.go
new file mode 100644
index 0000000..c7da74d
--- /dev/null
+++ b/internal/jobspec/dispatch_test.go
@@ -0,0 +1,222 @@
+package jobspec
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestDispatch_Markdown(t *testing.T) {
+	input := "---\nkind: Job\nname: md-job\n---\nbody content\n"
+	ws, err := Dispatch([]byte(input), "spec.md")
+	if err != nil {
+		t.Fatalf("Dispatch .md: %v", err)
+	}
+	if ws.Kind != "Job" {
+		t.Errorf("Kind = %q, want Job", ws.Kind)
+	}
+	if ws.Name != "md-job" {
+		t.Errorf("Name = %q, want md-job", ws.Name)
+	}
+	if ws.Body != "body content\n" {
+		t.Errorf("Body = %q, want %q (R-015)", ws.Body, "body content\n")
+	}
+}
+
+func TestDispatch_YAML(t *testing.T) {
+	input := "kind: Service\nname: yaml-svc\nports:\n  - name: http\n    port: 80\n"
+	ws, err := Dispatch([]byte(input), "spec.yaml")
+	if err != nil {
+		t.Fatalf("Dispatch .yaml: %v", err)
+	}
+	if ws.Kind != "Service" {
+		t.Errorf("Kind = %q, want Service", ws.Kind)
+	}
+	if ws.Name != "yaml-svc" {
+		t.Errorf("Name = %q, want yaml-svc", ws.Name)
+	}
+	if ws.Body != "" {
+		t.Errorf("Body = %q, want empty (YAML has no body)", ws.Body)
+	}
+	if len(ws.Ports) != 1 || ws.Ports[0].Name != "http" || ws.Ports[0].Port != 80 {
+		t.Errorf("Ports = %+v, want one http:80", ws.Ports)
+	}
+}
+
+func TestDispatch_YML(t *testing.T) {
+	input := "kind: DaemonSet\nname: yml-ds\n"
+	ws, err := Dispatch([]byte(input), "spec.yml")
+	if err != nil {
+		t.Fatalf("Dispatch .yml: %v", err)
+	}
+	if ws.Kind != "DaemonSet" {
+		t.Errorf("Kind = %q, want DaemonSet", ws.Kind)
+	}
+	if ws.Body != "" {
+		t.Errorf("Body = %q, want empty", ws.Body)
+	}
+}
+
+func TestDispatch_HCLAdapter(t *testing.T) {
+	hcl := `job "demo" {}
+task "build" {
+  command = "/bin/echo"
+  args    = ["hello"]
+}
+`
+	ws, err := Dispatch([]byte(hcl), "spec.hcl")
+	if err != nil {
+		t.Fatalf("Dispatch .hcl: %v", err)
+	}
+	if ws.Kind != "Job" {
+		t.Errorf("Kind = %q, want Job (adapter always sets Job)", ws.Kind)
+	}
+	if ws.Name != "demo" {
+		t.Errorf("Name = %q, want demo (from spec.Job.Name)", ws.Name)
+	}
+	if ws.Runtime == nil {
+		t.Fatal("Runtime is nil; adapter should populate from tasks[0]")
+	}
+	if ws.Runtime.OneOf != "process" {
+		t.Errorf("Runtime.OneOf = %q, want process", ws.Runtime.OneOf)
+	}
+	if ws.Runtime.Command != "/bin/echo" {
+		t.Errorf("Runtime.Command = %q, want /bin/echo (from tasks[0].Command)", ws.Runtime.Command)
+	}
+	if ws.Body != "" {
+		t.Errorf("Body = %q, want empty (HCL has no body)", ws.Body)
+	}
+}
+
+func TestDispatch_HCLAdapterNoTasks(t *testing.T) {
+	hcl := `job "x" {}`
+	_, err := Dispatch([]byte(hcl), "spec.hcl")
+	if err == nil {
+		t.Fatal("expected error for HCL with no tasks")
+	}
+	if !strings.Contains(err.Error(), "at least one task") {
+		t.Errorf("error = %q, want it to contain 'at least one task'", err.Error())
+	}
+}
+
+func TestDispatch_UnknownExtension(t *testing.T) {
+	_, err := Dispatch([]byte("kind: Job\nname: x\n"), "spec.json")
+	if err == nil {
+		t.Fatal("expected error for unknown extension, got nil")
+	}
+	if !strings.Contains(err.Error(), "unknown extension") {
+		t.Errorf("error = %q, want it to contain 'unknown extension'", err.Error())
+	}
+}
+
+func TestDispatch_NoExtension(t *testing.T) {
+	_, err := Dispatch([]byte("kind: Job\nname: x\n"), "spec")
+	if err == nil {
+		t.Fatal("expected error for no extension, got nil")
+	}
+}
+
+func TestDispatch_EmptyYAML(t *testing.T) {
+	_, err := Dispatch([]byte(""), "spec.yaml")
+	if err == nil {
+		t.Fatal("expected error for empty YAML, got nil")
+	}
+}
+
+func TestParseFile_Markdown(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "spec.md")
+	content := "---\nkind: Job\nname: file-md\n---\nbody\n"
+	if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	ws, err := ParseFile(path)
+	if err != nil {
+		t.Fatalf("ParseFile .md: %v", err)
+	}
+	if ws.Kind != "Job" || ws.Name != "file-md" {
+		t.Errorf("got Kind=%q Name=%q", ws.Kind, ws.Name)
+	}
+	if ws.Body != "body\n" {
+		t.Errorf("Body = %q, want %q", ws.Body, "body\n")
+	}
+}
+
+func TestParseFile_YAML(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "spec.yaml")
+	content := "kind: Service\nname: file-yaml\n"
+	if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	ws, err := ParseFile(path)
+	if err != nil {
+		t.Fatalf("ParseFile .yaml: %v", err)
+	}
+	if ws.Kind != "Service" || ws.Name != "file-yaml" {
+		t.Errorf("got Kind=%q Name=%q", ws.Kind, ws.Name)
+	}
+	if ws.Body != "" {
+		t.Errorf("Body = %q, want empty", ws.Body)
+	}
+}
+
+func TestParseFile_HCL(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "spec.hcl")
+	content := `job "file-hcl" {}
+task "t" { command = "/bin/true" }
+`
+	if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	ws, err := ParseFile(path)
+	if err != nil {
+		t.Fatalf("ParseFile .hcl: %v", err)
+	}
+	if ws.Kind != "Job" || ws.Name != "file-hcl" {
+		t.Errorf("got Kind=%q Name=%q", ws.Kind, ws.Name)
+	}
+	if ws.Runtime == nil || ws.Runtime.Command != "/bin/true" {
+		t.Errorf("Runtime.Command = %v, want /bin/true", ws.Runtime)
+	}
+}
+
+func TestParseFile_MissingFile(t *testing.T) {
+	_, err := ParseFile(filepath.Join(t.TempDir(), "nope.md"))
+	if err == nil {
+		t.Fatal("expected error for missing file, got nil")
+	}
+	if !strings.Contains(err.Error(), "read spec file") {
+		t.Errorf("error = %q, want it to contain 'read spec file'", err.Error())
+	}
+}
+
+func TestParseFile_UnknownExtension(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "spec.txt")
+	if err := os.WriteFile(path, []byte("kind: Job\nname: x\n"), 0o644); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	_, err := ParseFile(path)
+	if err == nil {
+		t.Fatal("expected error for unknown extension, got nil")
+	}
+	if !strings.Contains(err.Error(), "unknown extension") {
+		t.Errorf("error = %q, want 'unknown extension'", err.Error())
+	}
+}
+
+func TestParseHCL_LegacySpec(t *testing.T) {
+	hcl := `job "legacy" {}
+task "t" { command = "/bin/echo" }
+`
+	ws, err := ParseHCL([]byte(hcl), "spec.hcl")
+	if err != nil {
+		t.Fatalf("ParseHCL: %v", err)
+	}
+	if ws.Kind != "Job" || ws.Name != "legacy" {
+		t.Errorf("adapter got Kind=%q Name=%q", ws.Kind, ws.Name)
+	}
+}
diff --git a/internal/jobspec/markdown.go b/internal/jobspec/markdown.go
new file mode 100644
index 0000000..e1f346e
--- /dev/null
+++ b/internal/jobspec/markdown.go
@@ -0,0 +1,586 @@
+package jobspec
+
+import (
+	"fmt"
+	"strconv"
+	"strings"
+)
+
+// WorkloadSpec is the unified canonical jobspec populated by both the
+// Markdown frontmatter parser (canonical path, R-013/R-014) and the HCL
+// legacy adapter (REQ-064, REQ-090). It is the single shape consumed by
+// downstream phases (P0c schemas, P01 transport). The Markdown body
+// after the closing `---` is preserved verbatim in Body (R-015
+// byte-exact preservation is a load-bearing invariant enforced by the
+// fuzz harness in markdown_fuzz_test.go).
+type WorkloadSpec struct {
+	SpecVersion string
+	Kind        string
+	Name        string
+	Runtime     *RuntimeBlock
+	Count       int
+	Ports       []PortSpec
+	Env         map[string]string
+	Secrets     []string
+	Volumes     []VolumeSpec
+	Body        string
+}
+
+// RuntimeBlock is a minimal runtime abstraction surface populated by the
+// Markdown parser. The full runtime abstraction lands in P07; for now
+// only the one_of/image/command fields are parsed and stored (REQ-064).
+type RuntimeBlock struct {
+	OneOf   string
+	Image   string
+	Command string
+}
+
+// PortSpec is a minimal port binding entry. HostIP is optional.
+type PortSpec struct {
+	Name     string
+	HostPort int
+	Port     int
+	Protocol string
+	HostIP   string
+}
+
+// VolumeSpec is a minimal volume mount entry. Fields are stored raw
+// pending the P0c schema work (REQ-074).
+type VolumeSpec struct {
+	Name     string
+	Type     string
+	Source   string
+	Target   string
+	ReadOnly bool
+}
+
+// validKinds is the set of workload kinds accepted by the parser per
+// R-012. Unknown kinds are rejected.
+var validKinds = map[string]bool{
+	"Job":       true,
+	"Service":   true,
+	"DaemonSet": true,
+}
+
+// ParseMarkdown parses a Markdown jobspec with YAML frontmatter into a
+// *WorkloadSpec (R-013 canonical format, R-014 frontmatter). The body
+// after the closing `---` is preserved verbatim in result.Body
+// (R-015 byte-exact, including trailing newlines, CRLF, and BOM in the
+// body). The frontmatter parser is a minimal hand-rolled YAML-ish
+// key:value reader — gopkg.in/yaml.v3 is intentionally not added (same
+// approach as internal/config/markdown.go and internal/ns/parse.go).
+//
+// For .yaml/.yml files (no Markdown body), the dispatcher calls this
+// with the whole file treated as frontmatter and Body left empty (see
+// dispatch.go).
+func ParseMarkdown(data []byte) (*WorkloadSpec, error) {
+	content := string(data)
+	block, body, ok := splitFrontmatter(content)
+	if !ok {
+		return nil, fmt.Errorf("parse markdown: missing frontmatter delimiters")
+	}
+	if strings.TrimSpace(block) == "" {
+		return nil, fmt.Errorf("parse markdown: empty frontmatter")
+	}
+	spec, err := parseFrontmatterBlock(block)
+	if err != nil {
+		return nil, err
+	}
+	spec.Body = body
+	if err := validateWorkload(spec); err != nil {
+		return nil, err
+	}
+	return spec, nil
+}
+
+// splitFrontmatter splits the file content into the YAML frontmatter
+// block and the verbatim body that follows the closing `---`. A leading
+// UTF-8 BOM is stripped from the frontmatter scan (R-015: BOM is not
+// preserved in the frontmatter, but a BOM inside the body would be
+// preserved because the body is verbatim). Returns (block, body, ok).
+// ok is false when no opening `---` delimiter is present, or no closing
+// `---` delimiter is found, or the block is empty after the opening
+// delimiter (handled by caller).
+func splitFrontmatter(content string) (block, body string, ok bool) {
+	// Strip a leading UTF-8 BOM if present (EF BB BF). Only the
+	// frontmatter scan is BOM-stripped; the body is byte-exact, so a BOM
+	// appearing inside the body is preserved verbatim.
+	stripped := content
+	if strings.HasPrefix(stripped, "\uFEFF") {
+		stripped = stripped[len("\uFEFF"):]
+	}
+	// Trim leading horizontal whitespace and newlines before the
+	// opening delimiter. We do NOT trim trailing — body must be exact.
+	trimmed := strings.TrimLeft(stripped, "\r\n\t ")
+	if !strings.HasPrefix(trimmed, "---") {
+		return "", "", false
+	}
+	// The opening delimiter must be on its own line: `---` optionally
+	// followed by a line terminator.
+	rest := trimmed[3:]
+	// The opening `---` must be followed by a newline or end-of-file
+	// (a `---foo` prefix is not a valid delimiter).
+	if len(rest) > 0 && rest[0] != '\n' && rest[0] != '\r' {
+		return "", "", false
+	}
+	rest = strings.TrimLeft(rest, "\r\n")
+
+	// Find the closing delimiter line. The closing `---` must be on its
+	// own line: preceded by a newline (or at the start of `rest`) and
+	// followed by a newline or end-of-file.
+	idx := findClosingDelimiter(rest)
+	if idx < 0 {
+		return "", "", false
+	}
+	block = rest[:idx]
+	// Body is everything after the closing `---` line. The closing
+	// delimiter line itself (including its trailing newline) is NOT
+	// part of the body. We compute the byte offset in the original
+	// `content` so the body is byte-exact (R-015).
+	afterClose := rest[idx:]
+	// afterClose starts with `---`. Strip the delimiter line.
+	delimLen := 3
+	// Account for an optional trailing `...` or spaces on the delimiter
+	// line — the delimiter is `---` followed by anything up to and
+	// including the line terminator. Body starts after the newline.
+	// Find the end of the delimiter line.
+	newlineIdx := strings.IndexAny(afterClose, "\r\n")
+	var bodyStart int
+	if newlineIdx < 0 {
+		// Closing `---` is the last line: body is empty.
+		bodyStart = len(afterClose)
+	} else {
+		// Consume the delimiter line and its line terminator(s).
+		bodyStart = newlineIdx
+		// Strip a single CRLF or LF.
+		if strings.HasPrefix(afterClose[bodyStart:], "\r\n") {
+			bodyStart += 2
+		} else {
+			bodyStart += 1
+		}
+	}
+	body = afterClose[bodyStart:]
+	_ = delimLen
+	return block, body, true
+}
+
+// findClosingDelimiter returns the byte index in `rest` where the
+// closing `---` delimiter line begins, or -1 if none is found. The
+// delimiter must be on its own line: either at the start of `rest` or
+// preceded by a newline, and followed by a newline or end-of-file.
+func findClosingDelimiter(rest string) int {
+	// Special case: closing delimiter at the very start (frontmatter
+	// block is empty). The opening `---` is immediately followed by the
+	// closing `---`. We require the opening to be its own line, so the
+	// closing at index 0 means the opening had no body — invalid (empty
+	// frontmatter handled by caller). We still report it; caller
+	// rejects empty block.
+	for i := 0; i < len(rest); i++ {
+		if rest[i] != '\n' {
+			continue
+		}
+		// Candidate: line after this newline starts with `---`.
+		j := i + 1
+		if j+3 <= len(rest) && rest[j] == '-' && rest[j+1] == '-' && rest[j+2] == '-' {
+			// Must be followed by newline, CRLF, or end-of-file.
+			end := j + 3
+			if end == len(rest) {
+				return j
+			}
+			if rest[end] == '\n' || rest[end] == '\r' {
+				return j
+			}
+		}
+	}
+	// Final candidate: closing delimiter at the very start of rest
+	// (immediately after the opening delimiter + its newline). This
+	// happens when frontmatter is empty: `---\n---\n`. We already trim
+	// leading newlines off `rest`, so if rest itself starts with `---`
+	// AND it's a closing delimiter (followed by newline/EOF), it is the
+	// empty-frontmatter case.
+	if strings.HasPrefix(rest, "---") {
+		end := 3
+		if end == len(rest) {
+			return 0
+		}
+		if rest[end] == '\n' || rest[end] == '\r' {
+			return 0
+		}
+	}
+	return -1
+}
+
+// parseFrontmatterBlock parses a minimal YAML-ish frontmatter block into
+// a *WorkloadSpec (without Body, which is filled by the caller).
+//
+// Supported shapes:
+//
+//	kind: Job
+//	name: my-job
+//	count: 3
+//	runtime:
+//	  one_of: process
+//	  image: docker.io/nginx:latest
+//	  command: /bin/sh -c
+//	ports:
+//	  - name: http
+//	    port: 8080
+//	    host_port: 80
+//	    protocol: tcp
+//	env:
+//	  FOO: bar
+//	  BAR:
+//	    from: secret:my-secret
+//	secrets:
+//	  - db-password
+//	volumes:
+//	  - name: data
+//	    type: host
+//	    source: /data
+//	    target: /data
+//	    read_only: true
+//
+// Comments (# ...) and blank lines are ignored. Quoted scalar values
+// ("..." or '...') are unwrapped. No flow collections except the
+// inline-array form for `secrets`. Multi-line block scalars (|, >) are
+// not supported — by design, to avoid adding a YAML dependency for this
+// small surface.
+func parseFrontmatterBlock(block string) (*WorkloadSpec, error) {
+	spec := &WorkloadSpec{Count: 1}
+	lines := strings.Split(block, "\n")
+
+	type section int
+	const (
+		secNone section = iota
+		secRuntime
+		secPorts
+		secEnv
+		secSecrets
+		secVolumes
+	)
+	cur := secNone
+	var curPort *PortSpec
+	var curVol *VolumeSpec
+
+	flushPort := func() {
+		if curPort != nil {
+			spec.Ports = append(spec.Ports, *curPort)
+			curPort = nil
+		}
+	}
+	flushVol := func() {
+		if curVol != nil {
+			spec.Volumes = append(spec.Volumes, *curVol)
+			curVol = nil
+		}
+	}
+
+	for lineNo, raw := range lines {
+		line := stripComment(raw)
+		if strings.TrimSpace(line) == "" {
+			continue
+		}
+		indent := countIndent(line)
+		trimmed := strings.TrimSpace(line)
+
+		if indent == 0 {
+			// Flush any pending nested entry before switching sections.
+			flushPort()
+			flushVol()
+			cur = secNone
+
+			key, val, ok := splitKV(trimmed)
+			if !ok {
+				return nil, fmt.Errorf("parse markdown: line %d: malformed key:value", lineNo+1)
+			}
+			switch key {
+			case "orca-spec-version":
+				spec.SpecVersion = unquote(val)
+			case "kind":
+				spec.Kind = unquote(val)
+			case "name":
+				spec.Name = unquote(val)
+			case "count":
+				if n, err := strconv.Atoi(strings.TrimSpace(unquote(val))); err == nil {
+					spec.Count = n
+				} else {
+					return nil, fmt.Errorf("parse markdown: line %d: count: %v", lineNo+1, err)
+				}
+			case "runtime":
+				spec.Runtime = &RuntimeBlock{}
+				if strings.TrimSpace(val) != "" {
+					// Inline value (unusual); ignore — runtime is a block.
+				}
+				cur = secRuntime
+			case "ports":
+				cur = secPorts
+			case "env":
+				spec.Env = map[string]string{}
+				cur = secEnv
+			case "secrets":
+				if strings.TrimSpace(val) != "" {
+					arr, err := parseStringArray(val)
+					if err != nil {
+						return nil, fmt.Errorf("parse markdown: line %d: secrets: %w", lineNo+1, err)
+					}
+					spec.Secrets = append(spec.Secrets, arr...)
+					cur = secNone
+				} else {
+					cur = secSecrets
+				}
+			case "volumes":
+				cur = secVolumes
+			default:
+				// Unknown top-level keys are ignored (forward-compat).
+				cur = secNone
+			}
+			continue
+		}
+
+		// Indented line: a nested entry under the current section.
+		switch cur {
+		case secRuntime:
+			if spec.Runtime == nil {
+				spec.Runtime = &RuntimeBlock{}
+			}
+			key, val, ok := splitKV(trimmed)
+			if !ok {
+				continue
+			}
+			switch key {
+			case "one_of":
+				spec.Runtime.OneOf = unquote(val)
+			case "image":
+				spec.Runtime.Image = unquote(val)
+			case "command":
+				spec.Runtime.Command = unquote(val)
+			}
+		case secPorts:
+			if strings.HasPrefix(trimmed, "- ") || trimmed == "-" {
+				flushPort()
+				p := PortSpec{}
+				curPort = &p
+				rest := strings.TrimSpace(strings.TrimPrefix(trimmed, "-"))
+				if rest != "" {
+					applyPortKV(curPort, rest)
+				}
+			} else if curPort != nil {
+				applyPortKV(curPort, trimmed)
+			}
+		case secEnv:
+			key, val, ok := splitKV(trimmed)
+			if !ok {
+				continue
+			}
+			if val == "" {
+				// Nested mapping under env (e.g. `BAR:\n  from: ...`).
+				// Store the raw string for now (REQ-064: store raw).
+				spec.Env[key] = ""
+			} else if strings.HasPrefix(val, "{") && strings.HasSuffix(val, "}") {
+				// Inline object form: `BAR: {from: "secret:..."}`.
+				// Store the raw object string for now.
+				spec.Env[key] = val
+			} else {
+				spec.Env[key] = unquote(val)
+			}
+		case secSecrets:
+			if strings.HasPrefix(trimmed, "- ") || trimmed == "-" {
+				item := strings.TrimSpace(strings.TrimPrefix(trimmed, "-"))
+				if item != "" {
+					spec.Secrets = append(spec.Secrets, unquote(item))
+				}
+			}
+		case secVolumes:
+			if strings.HasPrefix(trimmed, "- ") || trimmed == "-" {
+				flushVol()
+				v := VolumeSpec{}
+				curVol = &v
+				rest := strings.TrimSpace(strings.TrimPrefix(trimmed, "-"))
+				if rest != "" {
+					applyVolumeKV(curVol, rest)
+				}
+			} else if curVol != nil {
+				applyVolumeKV(curVol, trimmed)
+			}
+		}
+	}
+	flushPort()
+	flushVol()
+	return spec, nil
+}
+
+// applyPortKV applies a `key: value` pair to a PortSpec entry.
+func applyPortKV(p *PortSpec, s string) {
+	key, val, ok := splitKV(s)
+	if !ok {
+		return
+	}
+	switch key {
+	case "name":
+		p.Name = unquote(val)
+	case "host_port":
+		if n, err := strconv.Atoi(strings.TrimSpace(unquote(val))); err == nil {
+			p.HostPort = n
+		}
+	case "port":
+		if n, err := strconv.Atoi(strings.TrimSpace(unquote(val))); err == nil {
+			p.Port = n
+		}
+	case "protocol":
+		p.Protocol = unquote(val)
+	case "host_ip":
+		p.HostIP = unquote(val)
+	}
+}
+
+// applyVolumeKV applies a `key: value` pair to a VolumeSpec entry.
+func applyVolumeKV(v *VolumeSpec, s string) {
+	key, val, ok := splitKV(s)
+	if !ok {
+		return
+	}
+	switch key {
+	case "name":
+		v.Name = unquote(val)
+	case "type":
+		v.Type = unquote(val)
+	case "source":
+		v.Source = unquote(val)
+	case "target":
+		v.Target = unquote(val)
+	case "read_only":
+		switch strings.ToLower(strings.TrimSpace(unquote(val))) {
+		case "true", "yes", "on", "1":
+			v.ReadOnly = true
+		}
+	}
+}
+
+// validateWorkload enforces required fields and kind validity (R-012).
+func validateWorkload(spec *WorkloadSpec) error {
+	if spec.Kind == "" {
+		return fmt.Errorf("parse markdown: missing kind")
+	}
+	if !validKinds[spec.Kind] {
+		return fmt.Errorf("parse markdown: kind %q is not one of Job, Service, DaemonSet", spec.Kind)
+	}
+	if strings.TrimSpace(spec.Name) == "" {
+		return fmt.Errorf("parse markdown: missing name")
+	}
+	return nil
+}
+
+// parseStringArray parses an inline YAML flow-array of scalars, e.g.
+// `["a", "b"]` or `['a', 'b']` or `[a, b]`. Empty array `[]` returns nil.
+func parseStringArray(val string) ([]string, error) {
+	val = strings.TrimSpace(val)
+	if val == "" {
+		return nil, nil
+	}
+	if !strings.HasPrefix(val, "[") || !strings.HasSuffix(val, "]") {
+		return nil, fmt.Errorf("expected [..] array, got %q", val)
+	}
+	inner := strings.TrimSpace(val[1 : len(val)-1])
+	if inner == "" {
+		return nil, nil
+	}
+	parts := splitFlowItems(inner)
+	out := make([]string, 0, len(parts))
+	for _, p := range parts {
+		p = strings.TrimSpace(p)
+		if p == "" {
+			continue
+		}
+		out = append(out, unquote(p))
+	}
+	return out, nil
+}
+
+// splitFlowItems splits a comma-separated flow-array body, respecting
+// single and double quotes.
+func splitFlowItems(s string) []string {
+	var out []string
+	inSingle := false
+	inDouble := false
+	start := 0
+	for i := 0; i < len(s); i++ {
+		c := s[i]
+		switch c {
+		case '\'':
+			if !inDouble {
+				inSingle = !inSingle
+			}
+		case '"':
+			if !inSingle {
+				inDouble = !inDouble
+			}
+		case ',':
+			if !inSingle && !inDouble {
+				out = append(out, s[start:i])
+				start = i + 1
+			}
+		}
+	}
+	out = append(out, s[start:])
+	return out
+}
+
+func countIndent(s string) int {
+	n := 0
+	for _, r := range s {
+		if r == ' ' || r == '\t' {
+			n++
+			continue
+		}
+		break
+	}
+	return n
+}
+
+func splitKV(s string) (key, val string, ok bool) {
+	idx := strings.Index(s, ":")
+	if idx < 0 {
+		return "", "", false
+	}
+	key = strings.TrimSpace(s[:idx])
+	val = strings.TrimSpace(s[idx+1:])
+	if key == "" {
+		return "", "", false
+	}
+	return key, val, true
+}
+
+func stripComment(s string) string {
+	inSingle := false
+	inDouble := false
+	for i := 0; i < len(s); i++ {
+		c := s[i]
+		switch c {
+		case '\'':
+			if !inDouble {
+				inSingle = !inSingle
+			}
+		case '"':
+			if !inSingle {
+				inDouble = !inDouble
+			}
+		case '#':
+			if !inSingle && !inDouble {
+				if i == 0 || s[i-1] == ' ' || s[i-1] == '\t' {
+					return s[:i]
+				}
+			}
+		}
+	}
+	return s
+}
+
+func unquote(s string) string {
+	s = strings.TrimSpace(s)
+	if len(s) >= 2 {
+		if (s[0] == '"' && s[len(s)-1] == '"') || (s[0] == '\'' && s[len(s)-1] == '\'') {
+			return s[1 : len(s)-1]
+		}
+	}
+	return s
+}
diff --git a/internal/jobspec/markdown_fuzz_test.go b/internal/jobspec/markdown_fuzz_test.go
new file mode 100644
index 0000000..71a991b
--- /dev/null
+++ b/internal/jobspec/markdown_fuzz_test.go
@@ -0,0 +1,109 @@
+package jobspec
+
+import (
+	"strings"
+	"testing"
+)
+
+// FuzzParseMarkdownRoundTrip is the REQ-067 fuzz harness for R-015
+// byte-exact body preservation. It generates random frontmatter + body
+// combinations, runs ParseMarkdown, and asserts that the parsed Body
+// equals the original body byte-for-byte whenever parsing succeeds.
+// When parsing fails (bad frontmatter), the iteration passes — the
+// parser is allowed to reject malformed input.
+//
+// The seed corpus (added via f.Add) covers adversarial fixtures: CRLF
+// body, BOM prefix, no frontmatter, only-closing-separator, body with
+// `---` inside a code fence, trailing whitespace, empty body. The seed
+// corpus runs as regular tests under `go test` (CI); random input runs
+// only under `go test -fuzz=FuzzParseMarkdownRoundTrip` in a dedicated
+// process.
+func FuzzParseMarkdownRoundTrip(f *testing.F) {
+	// Seed 1: valid frontmatter + simple body.
+	f.Add([]byte("---\nkind: Job\nname: seed1\n---\n# body\n"))
+
+	// Seed 2: CRLF body.
+	f.Add([]byte("---\r\nkind: Job\r\nname: seed2\r\n---\r\n# body\r\nCRLF\r\n"))
+
+	// Seed 3: BOM prefix.
+	f.Add([]byte("\uFEFF---\nkind: Job\nname: seed3\n---\nbody\n"))
+
+	// Seed 4: no frontmatter (just body) — should fail to parse.
+	f.Add([]byte("# just a body\nno frontmatter\n"))
+
+	// Seed 5: frontmatter with only the closing `---` (no opening).
+	f.Add([]byte("body\n---\nmore body\n"))
+
+	// Seed 6: body containing `---` in a code fence.
+	f.Add([]byte("---\nkind: Job\nname: seed6\n---\n```bash\necho '---'\n```\n"))
+
+	// Seed 7: body with trailing whitespace.
+	f.Add([]byte("---\nkind: Job\nname: seed7\n---\nbody with trailing spaces   \n"))
+
+	// Seed 8: empty body.
+	f.Add([]byte("---\nkind: Job\nname: seed8\n---\n"))
+
+	// Seed 9: empty frontmatter (should fail).
+	f.Add([]byte("---\n---\nbody\n"))
+
+	// Seed 10: body with no trailing newline.
+	f.Add([]byte("---\nkind: Job\nname: seed10\n---\nno trailing newline"))
+
+	f.Fuzz(func(t *testing.T, data []byte) {
+		// Reconstruct the body from the input so we can assert
+		// byte-exact round-trip. We do this by re-splitting the
+		// frontmatter using the same logic the parser uses, but only
+		// to extract the expected body. If the input has no valid
+		// frontmatter delimiter pair, ParseMarkdown will return an
+		// error and we pass the iteration.
+		expectedBody := extractExpectedBody(string(data))
+
+		spec, err := ParseMarkdown(data)
+		if err != nil {
+			// Parser rejected the input — acceptable for a fuzz
+			// iteration (the input may be malformed). Pass.
+			return
+		}
+		// R-015: body must be byte-exact.
+		if spec.Body != expectedBody {
+			t.Errorf("R-015 body round-trip mismatch:\n got = %q\nwant = %q", spec.Body, expectedBody)
+		}
+	})
+}
+
+// extractExpectedBody returns the body portion of a Markdown jobspec
+// input using the same delimiter-splitting logic as splitFrontmatter,
+// so the fuzz harness can assert byte-exact preservation independently
+// of the parser's internal extraction. If the input has no valid
+// frontmatter, the result is "" (and ParseMarkdown will error).
+func extractExpectedBody(content string) string {
+	stripped := content
+	if strings.HasPrefix(stripped, "\uFEFF") {
+		stripped = stripped[len("\uFEFF"):]
+	}
+	trimmed := strings.TrimLeft(stripped, "\r\n\t ")
+	if !strings.HasPrefix(trimmed, "---") {
+		return ""
+	}
+	rest := trimmed[3:]
+	if len(rest) > 0 && rest[0] != '\n' && rest[0] != '\r' {
+		return ""
+	}
+	rest = strings.TrimLeft(rest, "\r\n")
+	idx := findClosingDelimiter(rest)
+	if idx < 0 {
+		return ""
+	}
+	afterClose := rest[idx:]
+	newlineIdx := strings.IndexAny(afterClose, "\r\n")
+	if newlineIdx < 0 {
+		return ""
+	}
+	bodyStart := newlineIdx
+	if strings.HasPrefix(afterClose[bodyStart:], "\r\n") {
+		bodyStart += 2
+	} else {
+		bodyStart += 1
+	}
+	return afterClose[bodyStart:]
+}
diff --git a/internal/jobspec/markdown_test.go b/internal/jobspec/markdown_test.go
new file mode 100644
index 0000000..5d29ace
--- /dev/null
+++ b/internal/jobspec/markdown_test.go
@@ -0,0 +1,357 @@
+package jobspec
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestParseMarkdown_FullFrontmatter(t *testing.T) {
+	body := "# Hello\n\nThis is the body.\n\nTrailing newline preserved.\n"
+	input := "---\n" +
+		"orca-spec-version: \"1\"\n" +
+		"kind: Job\n" +
+		"name: my-job\n" +
+		"count: 3\n" +
+		"---\n" +
+		body
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if spec.SpecVersion != "1" {
+		t.Errorf("SpecVersion = %q, want %q", spec.SpecVersion, "1")
+	}
+	if spec.Kind != "Job" {
+		t.Errorf("Kind = %q, want %q", spec.Kind, "Job")
+	}
+	if spec.Name != "my-job" {
+		t.Errorf("Name = %q, want %q", spec.Name, "my-job")
+	}
+	if spec.Count != 3 {
+		t.Errorf("Count = %d, want 3", spec.Count)
+	}
+	if spec.Body != body {
+		t.Errorf("Body = %q, want %q (byte-exact, R-015)", spec.Body, body)
+	}
+}
+
+func TestParseMarkdown_BodyByteExactTrailingNewline(t *testing.T) {
+	cases := []struct {
+		name string
+		body string
+	}{
+		{"with_trailing_newline", "# Title\n\nbody\n"},
+		{"with_double_trailing_newline", "# Title\n\nbody\n\n"},
+		{"no_trailing_newline", "# Title\n\nbody"},
+		{"empty_body_with_newline", "\n"},
+		{"only_newlines", "\n\n\n"},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			input := "---\nkind: Job\nname: x\n---\n" + tc.body
+			spec, err := ParseMarkdown([]byte(input))
+			if err != nil {
+				t.Fatalf("ParseMarkdown: %v", err)
+			}
+			if spec.Body != tc.body {
+				t.Errorf("Body byte-exact mismatch (R-015):\n got = %q\nwant = %q", spec.Body, tc.body)
+			}
+		})
+	}
+}
+
+func TestParseMarkdown_NoFrontmatter(t *testing.T) {
+	input := "# Just a body\n\nNo frontmatter here."
+	_, err := ParseMarkdown([]byte(input))
+	if err == nil {
+		t.Fatal("expected error for missing frontmatter, got nil")
+	}
+	if !strings.Contains(err.Error(), "frontmatter") {
+		t.Errorf("error = %q, want it to contain 'frontmatter'", err.Error())
+	}
+}
+
+func TestParseMarkdown_EmptyFrontmatter(t *testing.T) {
+	input := "---\n---\n\nbody"
+	_, err := ParseMarkdown([]byte(input))
+	if err == nil {
+		t.Fatal("expected error for empty frontmatter, got nil")
+	}
+	if !strings.Contains(err.Error(), "empty frontmatter") {
+		t.Errorf("error = %q, want it to contain 'empty frontmatter'", err.Error())
+	}
+}
+
+func TestParseMarkdown_UnknownKind(t *testing.T) {
+	input := "---\nkind: CronJob\nname: x\n---\nbody\n"
+	_, err := ParseMarkdown([]byte(input))
+	if err == nil {
+		t.Fatal("expected error for unknown kind, got nil")
+	}
+	if !strings.Contains(err.Error(), "not one of") {
+		t.Errorf("error = %q, want it to contain 'not one of'", err.Error())
+	}
+}
+
+func TestParseMarkdown_MissingName(t *testing.T) {
+	input := "---\nkind: Job\n---\nbody\n"
+	_, err := ParseMarkdown([]byte(input))
+	if err == nil {
+		t.Fatal("expected error for missing name, got nil")
+	}
+	if !strings.Contains(err.Error(), "missing name") {
+		t.Errorf("error = %q, want it to contain 'missing name'", err.Error())
+	}
+}
+
+func TestParseMarkdown_MissingKind(t *testing.T) {
+	input := "---\nname: x\n---\nbody\n"
+	_, err := ParseMarkdown([]byte(input))
+	if err == nil {
+		t.Fatal("expected error for missing kind, got nil")
+	}
+	if !strings.Contains(err.Error(), "missing kind") {
+		t.Errorf("error = %q, want it to contain 'missing kind'", err.Error())
+	}
+}
+
+func TestParseMarkdown_EachValidKind(t *testing.T) {
+	cases := []string{"Job", "Service", "DaemonSet"}
+	for _, kind := range cases {
+		t.Run(kind, func(t *testing.T) {
+			input := "---\nkind: " + kind + "\nname: x\n---\nbody\n"
+			spec, err := ParseMarkdown([]byte(input))
+			if err != nil {
+				t.Fatalf("ParseMarkdown: %v", err)
+			}
+			if spec.Kind != kind {
+				t.Errorf("Kind = %q, want %q", spec.Kind, kind)
+			}
+		})
+	}
+}
+
+func TestParseMarkdown_EnvScalarAndObject(t *testing.T) {
+	input := "---\n" +
+		"kind: Job\n" +
+		"name: x\n" +
+		"env:\n" +
+		"  FOO: bar\n" +
+		"  BAZ: \"qux\"\n" +
+		"  SECRET_REF:\n" +
+		"    from: \"secret:db-password\"\n" +
+		"  INLINE: {from: \"secret:token\"}\n" +
+		"---\nbody\n"
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if got := spec.Env["FOO"]; got != "bar" {
+		t.Errorf("env[FOO] = %q, want %q", got, "bar")
+	}
+	if got := spec.Env["BAZ"]; got != "qux" {
+		t.Errorf("env[BAZ] = %q, want %q", got, "qux")
+	}
+	if got := spec.Env["INLINE"]; got != `{from: "secret:token"}` {
+		t.Errorf("env[INLINE] = %q, want the raw object string", got)
+	}
+	if _, ok := spec.Env["SECRET_REF"]; !ok {
+		t.Errorf("env[SECRET_REF] missing; nested from: stored as empty string")
+	}
+}
+
+func TestParseMarkdown_PortsArray(t *testing.T) {
+	input := "---\n" +
+		"kind: Service\n" +
+		"name: web\n" +
+		"ports:\n" +
+		"  - name: http\n" +
+		"    port: 8080\n" +
+		"    host_port: 80\n" +
+		"    protocol: tcp\n" +
+		"  - name: https\n" +
+		"    port: 8443\n" +
+		"    host_port: 443\n" +
+		"---\nbody\n"
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if len(spec.Ports) != 2 {
+		t.Fatalf("Ports = %d, want 2", len(spec.Ports))
+	}
+	if spec.Ports[0].Name != "http" || spec.Ports[0].Port != 8080 || spec.Ports[0].HostPort != 80 || spec.Ports[0].Protocol != "tcp" {
+		t.Errorf("Ports[0] = %+v", spec.Ports[0])
+	}
+	if spec.Ports[1].Name != "https" || spec.Ports[1].Port != 8443 || spec.Ports[1].HostPort != 443 {
+		t.Errorf("Ports[1] = %+v", spec.Ports[1])
+	}
+}
+
+func TestParseMarkdown_VolumesArray(t *testing.T) {
+	input := "---\n" +
+		"kind: Job\n" +
+		"name: x\n" +
+		"volumes:\n" +
+		"  - name: data\n" +
+		"    type: host\n" +
+		"    source: /data\n" +
+		"    target: /data\n" +
+		"    read_only: true\n" +
+		"---\nbody\n"
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if len(spec.Volumes) != 1 {
+		t.Fatalf("Volumes = %d, want 1", len(spec.Volumes))
+	}
+	v := spec.Volumes[0]
+	if v.Name != "data" || v.Type != "host" || v.Source != "/data" || v.Target != "/data" || !v.ReadOnly {
+		t.Errorf("Volumes[0] = %+v", v)
+	}
+}
+
+func TestParseMarkdown_RuntimeBlock(t *testing.T) {
+	input := "---\n" +
+		"kind: Job\n" +
+		"name: x\n" +
+		"runtime:\n" +
+		"  one_of: process\n" +
+		"  image: docker.io/nginx:latest\n" +
+		"  command: /bin/sh -c 'echo hi'\n" +
+		"---\nbody\n"
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if spec.Runtime == nil {
+		t.Fatal("Runtime is nil")
+	}
+	if spec.Runtime.OneOf != "process" {
+		t.Errorf("Runtime.OneOf = %q, want %q", spec.Runtime.OneOf, "process")
+	}
+	if spec.Runtime.Image != "docker.io/nginx:latest" {
+		t.Errorf("Runtime.Image = %q, want %q", spec.Runtime.Image, "docker.io/nginx:latest")
+	}
+	if spec.Runtime.Command != "/bin/sh -c 'echo hi'" {
+		t.Errorf("Runtime.Command = %q, want %q", spec.Runtime.Command, "/bin/sh -c 'echo hi'")
+	}
+}
+
+func TestParseMarkdown_SecretsInlineArray(t *testing.T) {
+	input := "---\nkind: Job\nname: x\nsecrets: [\"db-password\", \"api-token\"]\n---\nbody\n"
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if len(spec.Secrets) != 2 {
+		t.Fatalf("Secrets = %d, want 2", len(spec.Secrets))
+	}
+	if spec.Secrets[0] != "db-password" || spec.Secrets[1] != "api-token" {
+		t.Errorf("Secrets = %v, want [db-password api-token]", spec.Secrets)
+	}
+}
+
+func TestParseMarkdown_SecretsBlockArray(t *testing.T) {
+	input := "---\n" +
+		"kind: Job\n" +
+		"name: x\n" +
+		"secrets:\n" +
+		"  - db-password\n" +
+		"  - api-token\n" +
+		"---\nbody\n"
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if len(spec.Secrets) != 2 {
+		t.Fatalf("Secrets = %d, want 2", len(spec.Secrets))
+	}
+	if spec.Secrets[0] != "db-password" || spec.Secrets[1] != "api-token" {
+		t.Errorf("Secrets = %v, want [db-password api-token]", spec.Secrets)
+	}
+}
+
+func TestParseMarkdown_CRLFBodyPreserved(t *testing.T) {
+	body := "# Title\r\n\r\nCRLF body.\r\n"
+	input := "---\r\nkind: Job\r\nname: x\r\n---\r\n" + body
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if spec.Body != body {
+		t.Errorf("CRLF body not preserved (R-015):\n got = %q\nwant = %q", spec.Body, body)
+	}
+}
+
+func TestParseMarkdown_BOMStrippedFromFrontmatter(t *testing.T) {
+	body := "# body\n"
+	input := "\uFEFF" + "---\nkind: Job\nname: x\n---\n" + body
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if spec.Kind != "Job" {
+		t.Errorf("Kind = %q, want Job (BOM should be stripped from frontmatter scan)", spec.Kind)
+	}
+	if spec.Body != body {
+		t.Errorf("Body = %q, want %q", spec.Body, body)
+	}
+}
+
+func TestParseMarkdown_BodyWithCodeFenceContainingDashes(t *testing.T) {
+	body := "```bash\n" +
+		"echo '---'\n" +
+		"echo '--- end ---'\n" +
+		"```\n"
+	input := "---\nkind: Job\nname: x\n---\n" + body
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if spec.Body != body {
+		t.Errorf("Body with code-fence --- not preserved (R-015):\n got = %q\nwant = %q", spec.Body, body)
+	}
+}
+
+func TestParseMarkdown_OnlyClosingSeparator(t *testing.T) {
+	input := "no opening\n---\nbody\n"
+	_, err := ParseMarkdown([]byte(input))
+	if err == nil {
+		t.Fatal("expected error for input with only closing separator, got nil")
+	}
+}
+
+func TestParseMarkdown_QuotedValues(t *testing.T) {
+	input := "---\nkind: \"Job\"\nname: 'my-job'\n---\nbody\n"
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if spec.Kind != "Job" {
+		t.Errorf("Kind = %q, want Job (double-quoted)", spec.Kind)
+	}
+	if spec.Name != "my-job" {
+		t.Errorf("Name = %q, want my-job (single-quoted)", spec.Name)
+	}
+}
+
+func TestParseMarkdown_CountDefault(t *testing.T) {
+	input := "---\nkind: Job\nname: x\n---\nbody\n"
+	spec, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown: %v", err)
+	}
+	if spec.Count != 1 {
+		t.Errorf("Count default = %d, want 1", spec.Count)
+	}
+}
+
+func TestParseMarkdown_UnknownKeyIgnored(t *testing.T) {
+	input := "---\nkind: Job\nname: x\nfuture_field: value\n---\nbody\n"
+	_, err := ParseMarkdown([]byte(input))
+	if err != nil {
+		t.Fatalf("ParseMarkdown should ignore unknown keys: %v", err)
+	}
+}
diff --git a/internal/jobspec/spec.go b/internal/jobspec/spec.go
index 4cfa80e..be0bb35 100644
--- a/internal/jobspec/spec.go
+++ b/internal/jobspec/spec.go
@@ -2,7 +2,6 @@ package jobspec
 
 import (
 	"fmt"
-	"os"
 	"strings"
 
 	"github.com/hashicorp/hcl/v2"
@@ -10,16 +9,24 @@ import (
 	"github.com/hashicorp/hcl/v2/hclsimple"
 )
 
+// Spec is the legacy HCL-only jobspec shape. It is retained for the
+// v0.9→v0.10 migration window (REQ-090) and is populated by ParseHCLLegacy.
+//
+// Deprecated: HCL is legacy per R-013; new code should consume the
+// unified *WorkloadSpec returned by ParseFile/Dispatch (see
+// dispatch.go and markdown.go).
 type Spec struct {
 	Job   JobSpec    `hcl:"job,block"`
 	Tasks []TaskSpec `hcl:"task,block"`
 }
 
+// JobSpec is the legacy HCL job block.
 type JobSpec struct {
 	Name string `hcl:"name,label"`
 	Type string `hcl:"type,optional"`
 }
 
+// TaskSpec is the legacy HCL task block.
 type TaskSpec struct {
 	Name    string   `hcl:"name,label"`
 	Command string   `hcl:"command"`
@@ -27,34 +34,14 @@ type TaskSpec struct {
 	Env     []string `hcl:"env,optional"`
 }
 
-func ParseFile(path string) (*Spec, error) {
-	data, err := os.ReadFile(path)
-	if err != nil {
-		return nil, fmt.Errorf("read spec file: %w", err)
-	}
-	return Parse(data, path)
-}
-
-func Parse(data []byte, filename string) (*Spec, error) {
-	var spec Spec
-	err := hclsimple.Decode(filename, data, nil, &spec)
-	if err != nil {
-		return nil, fmt.Errorf("decode hcl: %w", err)
-	}
-	if spec.Job.Name == "" {
-		return nil, fmt.Errorf("spec missing job name")
-	}
-	if len(spec.Tasks) == 0 {
-		return nil, fmt.Errorf("spec must have at least one task")
-	}
-	for i, t := range spec.Tasks {
-		if t.Command == "" {
-			return nil, fmt.Errorf("task[%d] (%s) missing command", i, t.Name)
-		}
-	}
-	return &spec, nil
+// hclDecode wraps hclsimple.Decode for testability.
+func hclDecode(filename string, data []byte, spec *Spec) error {
+	return hclsimple.Decode(filename, data, nil, spec)
 }
 
+// Validate is the legacy HCL Spec validator retained for the migration
+// window (REQ-090). New code should use validateWorkload on a
+// *WorkloadSpec.
 func (s *Spec) Validate() error {
 	if strings.TrimSpace(s.Job.Name) == "" {
 		return fmt.Errorf("job name is required")
@@ -65,5 +52,17 @@ func (s *Spec) Validate() error {
 	return nil
 }
 
+// Parse is the original HCL-only entry point retained for backward
+// compatibility with direct HCL callers during the v0.9→v0.10 migration
+// window (REQ-090). New code should call the dispatcher ParseFile (which
+// returns *WorkloadSpec) or ParseHCL (which adapts HCL into
+// *WorkloadSpec).
+//
+// Deprecated: use ParseFile (dispatcher) or ParseHCL (adapter). HCL is
+// legacy per R-013.
+func Parse(data []byte, filename string) (*Spec, error) {
+	return ParseHCLLegacy(data, filename)
+}
+
 var _ = hcl.Diagnostics{}
 var _ = gohcl.DecodeBody
diff --git a/internal/jobspec/spec_test.go b/internal/jobspec/spec_test.go
index d1f0a3e..9502997 100644
--- a/internal/jobspec/spec_test.go
+++ b/internal/jobspec/spec_test.go
@@ -130,9 +130,9 @@ func TestParse_GoldenFiles(t *testing.T) {
 	for _, tc := range cases {
 		t.Run(tc.name, func(t *testing.T) {
 			path := filepath.Join("testdata", tc.file)
-			spec, err := ParseFile(path)
+			spec, err := ParseHCLFile(path)
 			if err != nil {
-				t.Fatalf("ParseFile(%s): %v", tc.file, err)
+				t.Fatalf("ParseHCLFile(%s): %v", tc.file, err)
 			}
 			if spec.Job.Name != tc.wantJob {
 				t.Errorf("job name = %q, want %q", spec.Job.Name, tc.wantJob)
@@ -254,9 +254,9 @@ func TestSpec_Validate(t *testing.T) {
 
 func TestSpec_Validate_RoundTripFromParse(t *testing.T) {
 	path := filepath.Join("testdata", "valid_single_task.hcl")
-	spec, err := ParseFile(path)
+	spec, err := ParseHCLFile(path)
 	if err != nil {
-		t.Fatalf("ParseFile: %v", err)
+		t.Fatalf("ParseHCLFile: %v", err)
 	}
 	if err := spec.Validate(); err != nil {
 		t.Errorf("Validate on parsed spec: %v", err)