diff --git a/roboco/foundation/policy/content/__init__.py b/roboco/foundation/policy/content/__init__.py
new file mode 100644
index 00000000..5393b79d
--- /dev/null
+++ b/roboco/foundation/policy/content/__init__.py
@@ -0,0 +1,45 @@
+"""Structured agent-content schema — the RoboCo content standard.
+
+Public surface: the content models, the ``CONTENT_MODELS`` registry, the
+``validate_content`` entry point, and the ``ContentValidationError`` raised on
+failure. See :mod:`.models` for the per-type schemas.
+"""
+
+from __future__ import annotations
+
+from .enums import Severity, Verdict
+from .models import (
+    CONTENT_MODELS,
+    AcVerdict,
+    AuditorNote,
+    DeveloperNote,
+    DocNote,
+    Finding,
+    PrReviewContent,
+    QaNote,
+    ResumptionNote,
+    TaskDescription,
+    WorkUnit,
+    required_shape,
+    validate_content,
+)
+from .validators import ContentValidationError
+
+__all__ = [
+    "CONTENT_MODELS",
+    "AcVerdict",
+    "AuditorNote",
+    "ContentValidationError",
+    "DeveloperNote",
+    "DocNote",
+    "Finding",
+    "PrReviewContent",
+    "QaNote",
+    "ResumptionNote",
+    "Severity",
+    "TaskDescription",
+    "Verdict",
+    "WorkUnit",
+    "required_shape",
+    "validate_content",
+]
diff --git a/roboco/foundation/policy/content/enums.py b/roboco/foundation/policy/content/enums.py
new file mode 100644
index 00000000..b4a80841
--- /dev/null
+++ b/roboco/foundation/policy/content/enums.py
@@ -0,0 +1,35 @@
+"""Enumerations for structured agent content.
+
+These are the controlled vocabularies the content schema models
+(:mod:`roboco.foundation.policy.content.models`) validate against. The cell
+``Team`` vocabulary is reused from :mod:`roboco.foundation.identity` rather
+than redefined here.
+"""
+
+from __future__ import annotations
+
+from enum import StrEnum
+
+
+class Severity(StrEnum):
+    """Severity ladder for a PR-review finding (worst → least)."""
+
+    BLOCKER = "blocker"  # must fix before merge
+    MAJOR = "major"  # significant defect
+    MINOR = "minor"  # small defect, fix advised
+    NIT = "nit"  # cosmetic / preference
+
+
+class Verdict(StrEnum):
+    """Outcome of a review.
+
+    ``approved`` / ``changes_requested`` describe a PR review; ``passed`` /
+    ``failed`` describe a QA or in-path gate review. All four are valid
+    ``PrReviewContent`` verdicts; ``QaNote`` accepts only ``passed`` /
+    ``failed``.
+    """
+
+    APPROVED = "approved"
+    CHANGES_REQUESTED = "changes_requested"
+    PASSED = "passed"
+    FAILED = "failed"
diff --git a/roboco/foundation/policy/content/models.py b/roboco/foundation/policy/content/models.py
new file mode 100644
index 00000000..23051b47
--- /dev/null
+++ b/roboco/foundation/policy/content/models.py
@@ -0,0 +1,447 @@
+"""Structured content schema models.
+
+Every human-facing artifact an agent produces — PR-review comments, task
+notes, task descriptions, resumption context — is one of these typed models.
+Each model:
+
+- validates its named fields (non-trivial, required, controlled vocab);
+- coerces a lone scalar where a list is declared (graceful, never a hard
+  reject of the well-intentioned single-item case);
+- renders to canonical labeled markdown via ``render_markdown()``.
+
+The structured payload is the source of truth; the rendered markdown is the
+derived mirror written to the Task's TEXT note columns and to PR comment
+bodies. ``CONTENT_MODELS`` maps the gateway's content-type key to its model;
+``validate_content`` is the single entry point that turns a raw payload into a
+validated model (or a ``ContentValidationError``).
+"""
+
+from __future__ import annotations
+
+from typing import Any, Literal
+
+from pydantic import (
+    BaseModel,
+    ConfigDict,
+    Field,
+    ValidationError,
+    ValidationInfo,
+    field_validator,
+    model_validator,
+)
+
+from roboco.foundation.identity import CELL_TEAMS, Team
+
+from .enums import Severity, Verdict
+from .validators import ContentValidationError, coerce_to_list, reject_trivial
+
+_SUMMARY_MIN = 10
+
+
+class _Base(BaseModel):
+    """Shared config: drop unknown keys (graceful), validate assignment."""
+
+    model_config = ConfigDict(extra="ignore", validate_assignment=True)
+
+
+class _Content(_Base):
+    """A top-level content type that renders to canonical markdown."""
+
+    def render_markdown(self) -> str:  # pragma: no cover - overridden
+        raise NotImplementedError
+
+
+# --------------------------------------------------------------------------- #
+# Markdown helpers
+# --------------------------------------------------------------------------- #
+
+
+def _bullets(items: list[str]) -> str:
+    return "\n".join(f"- {i.strip()}" for i in items if i and i.strip())
+
+
+def _section(title: str, body: str) -> str:
+    body = body.strip()
+    return f"## {title}\n{body}" if body else ""
+
+
+def _join(parts: list[str]) -> str:
+    return "\n\n".join(p for p in parts if p).strip()
+
+
+# --------------------------------------------------------------------------- #
+# Sub-models
+# --------------------------------------------------------------------------- #
+
+
+class Finding(_Base):
+    """One PR-review finding — file + line + expected vs actual."""
+
+    file: str
+    line: int | None = None
+    severity: Severity
+    criterion: str | None = None
+    expected: str
+    actual: str
+
+    @field_validator("file", "expected", "actual")
+    @classmethod
+    def _nontrivial(cls, v: str, info: ValidationInfo) -> str:
+        return reject_trivial(v, field=info.field_name or "field")
+
+
+class WorkUnit(_Base):
+    """One cell's slice of a task description."""
+
+    team: Team
+    summary: str
+    items: list[str] = Field(default_factory=list)
+
+    @field_validator("items", mode="before")
+    @classmethod
+    def _coerce_items(cls, v: Any) -> Any:
+        return coerce_to_list(v)
+
+    @field_validator("team")
+    @classmethod
+    def _cell_team_only(cls, v: Team) -> Team:
+        if v not in CELL_TEAMS:
+            raise ValueError(
+                f"team must be a cell ({', '.join(t.value for t in CELL_TEAMS)})"
+            )
+        return v
+
+    @field_validator("summary")
+    @classmethod
+    def _nontrivial_summary(cls, v: str) -> str:
+        return reject_trivial(v, field="summary")
+
+    @field_validator("items")
+    @classmethod
+    def _nonempty_items(cls, v: list[str]) -> list[str]:
+        cleaned = [i for i in v if i and i.strip()]
+        if not cleaned:
+            raise ValueError("items must list at least one work item")
+        return cleaned
+
+
+class AcVerdict(_Base):
+    """One acceptance-criterion verdict from a QA review."""
+
+    criterion: str
+    status: Literal["verified", "failed", "na"]
+    how: str
+
+    @field_validator("criterion", "how")
+    @classmethod
+    def _nontrivial(cls, v: str, info: ValidationInfo) -> str:
+        return reject_trivial(v, field=info.field_name or "field")
+
+
+# --------------------------------------------------------------------------- #
+# Top-level content models
+# --------------------------------------------------------------------------- #
+
+
+class PrReviewContent(_Content):
+    """A PR-review comment / reviewer verdict."""
+
+    summary: str
+    findings: list[Finding] = Field(default_factory=list)
+    verdict: Verdict
+
+    @field_validator("findings", mode="before")
+    @classmethod
+    def _coerce_findings(cls, v: Any) -> Any:
+        return coerce_to_list(v)
+
+    @field_validator("summary")
+    @classmethod
+    def _nontrivial_summary(cls, v: str) -> str:
+        return reject_trivial(v, field="summary", min_chars=_SUMMARY_MIN)
+
+    @model_validator(mode="after")
+    def _findings_required_for_negative(self) -> PrReviewContent:
+        if self.verdict in (Verdict.CHANGES_REQUESTED, Verdict.FAILED) and not (
+            self.findings
+        ):
+            raise ValueError(
+                "findings must be non-empty when verdict is changes_requested or failed"
+            )
+        return self
+
+    def render_markdown(self) -> str:
+        parts = [_section("Summary", self.summary)]
+        if self.findings:
+            rows = [
+                "| File | Line | Severity | Expected → Actual |",
+                "| --- | --- | --- | --- |",
+            ]
+            for f in self.findings:
+                loc = str(f.line) if f.line is not None else "—"
+                crit = f" ({f.criterion})" if f.criterion else ""
+                rows.append(
+                    f"| `{f.file}`{crit} | {loc} | {f.severity.value} "
+                    f"| {f.expected} → {f.actual} |"
+                )
+            parts.append("## Findings\n" + "\n".join(rows))
+        parts.append(_section("Verdict", self.verdict.value.replace("_", " ")))
+        return _join(parts)
+
+
+class TaskDescription(_Content):
+    """A well-formed task description (shared by PM delegate + Intake draft)."""
+
+    objective: str
+    what_this_builds: list[str] = Field(default_factory=list)
+    the_work: list[WorkUnit] = Field(default_factory=list)
+    notes: list[str] = Field(default_factory=list)
+    acceptance_criteria: list[str] = Field(default_factory=list)
+
+    @field_validator(
+        "what_this_builds",
+        "the_work",
+        "notes",
+        "acceptance_criteria",
+        mode="before",
+    )
+    @classmethod
+    def _coerce(cls, v: Any) -> Any:
+        return coerce_to_list(v)
+
+    @field_validator("objective")
+    @classmethod
+    def _nontrivial_objective(cls, v: str) -> str:
+        return reject_trivial(v, field="objective", min_chars=_SUMMARY_MIN)
+
+    @field_validator("the_work")
+    @classmethod
+    def _nonempty_work(cls, v: list[WorkUnit]) -> list[WorkUnit]:
+        if not v:
+            raise ValueError("the_work must contain at least one work unit")
+        return v
+
+    @field_validator("acceptance_criteria")
+    @classmethod
+    def _nonempty_ac(cls, v: list[str]) -> list[str]:
+        cleaned = [i for i in v if i and i.strip()]
+        if not cleaned:
+            raise ValueError("acceptance_criteria must list at least one criterion")
+        return cleaned
+
+    def render_markdown(self) -> str:
+        parts = [_section("Objective", self.objective)]
+        if self.what_this_builds:
+            parts.append(_section("What This Builds", _bullets(self.what_this_builds)))
+        if self.the_work:
+            units = []
+            for u in self.the_work:
+                head = f"**{u.team.value.replace('_', ' ').title()}** — {u.summary}"
+                units.append(f"{head}\n{_bullets(u.items)}")
+            parts.append("## The Work\n" + "\n\n".join(units))
+        if self.notes:
+            parts.append(_section("Notes", _bullets(self.notes)))
+        parts.append(
+            _section("Acceptance Criteria", _bullets(self.acceptance_criteria))
+        )
+        return _join(parts)
+
+
+class ResumptionNote(_Content):
+    """The human handoff that lives in ``quick_context`` (no machine markers)."""
+
+    done: str
+    next: str
+    where_to_look: list[str] = Field(default_factory=list)
+
+    @field_validator("where_to_look", mode="before")
+    @classmethod
+    def _coerce(cls, v: Any) -> Any:
+        return coerce_to_list(v)
+
+    @field_validator("done", "next")
+    @classmethod
+    def _nontrivial(cls, v: str, info: ValidationInfo) -> str:
+        return reject_trivial(v, field=info.field_name or "field")
+
+    def render_markdown(self) -> str:
+        parts = [_section("Done", self.done), _section("Next", self.next)]
+        if self.where_to_look:
+            parts.append(_section("Where to look", _bullets(self.where_to_look)))
+        return _join(parts)
+
+
+class DeveloperNote(_Content):
+    """A developer's task note."""
+
+    summary: str
+    changes: list[str] = Field(default_factory=list)
+    risks: list[str] = Field(default_factory=list)
+    follow_ups: list[str] = Field(default_factory=list)
+
+    @field_validator("changes", "risks", "follow_ups", mode="before")
+    @classmethod
+    def _coerce(cls, v: Any) -> Any:
+        return coerce_to_list(v)
+
+    @field_validator("summary")
+    @classmethod
+    def _nontrivial(cls, v: str) -> str:
+        return reject_trivial(v, field="summary", min_chars=_SUMMARY_MIN)
+
+    def render_markdown(self) -> str:
+        parts = [_section("Summary", self.summary)]
+        if self.changes:
+            parts.append(_section("Changes", _bullets(self.changes)))
+        if self.risks:
+            parts.append(_section("Risks", _bullets(self.risks)))
+        if self.follow_ups:
+            parts.append(_section("Follow-ups", _bullets(self.follow_ups)))
+        return _join(parts)
+
+
+class QaNote(_Content):
+    """A QA review note — summary + per-criterion verdicts + outcome."""
+
+    summary: str
+    ac_verdicts: list[AcVerdict] = Field(default_factory=list)
+    verdict: Verdict
+
+    @field_validator("ac_verdicts", mode="before")
+    @classmethod
+    def _coerce(cls, v: Any) -> Any:
+        return coerce_to_list(v)
+
+    @field_validator("summary")
+    @classmethod
+    def _nontrivial(cls, v: str) -> str:
+        return reject_trivial(v, field="summary", min_chars=_SUMMARY_MIN)
+
+    @field_validator("verdict")
+    @classmethod
+    def _passed_or_failed(cls, v: Verdict) -> Verdict:
+        if v not in (Verdict.PASSED, Verdict.FAILED):
+            raise ValueError("QA verdict must be 'passed' or 'failed'")
+        return v
+
+    @field_validator("ac_verdicts")
+    @classmethod
+    def _nonempty(cls, v: list[AcVerdict]) -> list[AcVerdict]:
+        if not v:
+            raise ValueError("ac_verdicts must cover at least one acceptance criterion")
+        return v
+
+    def render_markdown(self) -> str:
+        parts = [_section("Summary", self.summary)]
+        marks = {"verified": "✅", "failed": "❌", "na": "—"}
+        rows = [
+            f"- {marks[a.status]} **{a.criterion}** — {a.how}" for a in self.ac_verdicts
+        ]
+        parts.append("## Acceptance Criteria\n" + "\n".join(rows))
+        parts.append(_section("Verdict", self.verdict.value))
+        return _join(parts)
+
+
+class DocNote(_Content):
+    """A documenter's note — what was documented vs deliberately skipped."""
+
+    summary: str
+    documented: list[str] = Field(default_factory=list)
+    skipped: list[str] = Field(default_factory=list)
+
+    @field_validator("documented", "skipped", mode="before")
+    @classmethod
+    def _coerce(cls, v: Any) -> Any:
+        return coerce_to_list(v)
+
+    @field_validator("summary")
+    @classmethod
+    def _nontrivial(cls, v: str) -> str:
+        return reject_trivial(v, field="summary", min_chars=_SUMMARY_MIN)
+
+    def render_markdown(self) -> str:
+        parts = [_section("Summary", self.summary)]
+        if self.documented:
+            parts.append(_section("Documented", _bullets(self.documented)))
+        if self.skipped:
+            parts.append(_section("Skipped", _bullets(self.skipped)))
+        return _join(parts)
+
+
+class AuditorNote(_Content):
+    """An auditor's confidential observation."""
+
+    summary: str
+    concerns: list[str] = Field(default_factory=list)
+    severity: Literal["info", "watch", "risk"]
+
+    @field_validator("concerns", mode="before")
+    @classmethod
+    def _coerce(cls, v: Any) -> Any:
+        return coerce_to_list(v)
+
+    @field_validator("summary")
+    @classmethod
+    def _nontrivial(cls, v: str) -> str:
+        return reject_trivial(v, field="summary", min_chars=_SUMMARY_MIN)
+
+    def render_markdown(self) -> str:
+        parts = [
+            _section("Summary", self.summary),
+            _section("Severity", self.severity),
+        ]
+        if self.concerns:
+            parts.append(_section("Concerns", _bullets(self.concerns)))
+        return _join(parts)
+
+
+# --------------------------------------------------------------------------- #
+# Registry + single validation entry point
+# --------------------------------------------------------------------------- #
+
+CONTENT_MODELS: dict[str, type[_Content]] = {
+    "pr_review": PrReviewContent,
+    "task_description": TaskDescription,
+    "resumption": ResumptionNote,
+    "developer": DeveloperNote,
+    "qa": QaNote,
+    "doc": DocNote,
+    "auditor": AuditorNote,
+}
+
+
+def validate_content(content_type: str, payload: Any) -> _Content:
+    """Validate ``payload`` against the model for ``content_type``.
+
+    Returns the validated model (passed through unchanged if it is already an
+    instance of the right model). Raises ``ContentValidationError(field,
+    reason)`` — built from Pydantic's first error — on any failure, including an
+    unknown content type.
+    """
+    model_cls = CONTENT_MODELS.get(content_type)
+    if model_cls is None:
+        raise ContentValidationError(
+            "content_type", f"unknown content type: {content_type!r}"
+        )
+    if isinstance(payload, model_cls):
+        return payload
+    try:
+        return model_cls.model_validate(
+            payload if isinstance(payload, dict) else dict(payload or {})
+        )
+    except ValidationError as exc:
+        first = exc.errors()[0]
+        loc = ".".join(str(p) for p in first.get("loc", ())) or "?"
+        raise ContentValidationError(loc, first.get("msg", "invalid")) from exc
+
+
+def required_shape(content_type: str) -> dict[str, str]:
+    """A ``{field: type-hint}`` map for a content type, for remediation hints."""
+    model_cls = CONTENT_MODELS.get(content_type)
+    if model_cls is None:
+        return {}
+    shape: dict[str, str] = {}
+    for name, field in model_cls.model_fields.items():
+        ann = field.annotation
+        shape[name] = getattr(ann, "__name__", str(ann))
+    return shape
diff --git a/roboco/foundation/policy/content/validators.py b/roboco/foundation/policy/content/validators.py
new file mode 100644
index 00000000..afad25e8
--- /dev/null
+++ b/roboco/foundation/policy/content/validators.py
@@ -0,0 +1,85 @@
+"""Shared validation primitives for structured agent content.
+
+Two jobs:
+
+- ``reject_trivial`` — the non-empty / non-placeholder gate, reused by every
+  content model's field validators. Raises ``ValueError`` so it composes with
+  Pydantic field validators (which collect ``ValueError`` into a
+  ``ValidationError``).
+- ``ContentValidationError`` — the public, gateway-facing exception. The
+  top-level ``validate_content`` (in :mod:`.models`) converts Pydantic's
+  ``ValidationError`` into this so callers get a single ``(field, reason)``
+  shape to build a remediation envelope from.
+"""
+
+from __future__ import annotations
+
+from typing import Any
+
+# Placeholder tokens that are never an acceptable whole-field value. Extends the
+# commit-validator's banned single-word list (services/gateway/commit_validator).
+BANNED_PHRASES: frozenset[str] = frozenset(
+    {
+        "wip",
+        "tmp",
+        "tbd",
+        "todo",
+        "asdf",
+        "oops",
+        "stuff",
+        "things",
+        "n/a",
+        "na",
+        "none",
+        "null",
+        "-",
+        "--",
+        "...",
+        ".",
+        "x",
+    }
+)
+
+
+class ContentValidationError(Exception):
+    """A structured-content payload failed validation.
+
+    Carries the offending ``field`` and a human ``reason`` so the gateway can
+    return a remediable envelope (``{error, message, remediate, missing}``).
+    """
+
+    def __init__(self, field: str, reason: str) -> None:
+        self.field = field
+        self.reason = reason
+        super().__init__(f"{field}: {reason}")
+
+
+def reject_trivial(value: str, *, field: str, min_chars: int = 1) -> str:
+    """Return the trimmed value, or raise ``ValueError`` if it is trivial.
+
+    Trivial = empty, shorter than ``min_chars``, or a known placeholder token.
+    Raises ``ValueError`` (not ``ContentValidationError``) so it can be used
+    directly inside Pydantic field validators.
+    """
+    text = (value or "").strip()
+    if not text:
+        raise ValueError(f"{field} must not be empty")
+    if len(text) < min_chars:
+        raise ValueError(f"{field} must be at least {min_chars} characters")
+    if text.lower() in BANNED_PHRASES:
+        raise ValueError(f"{field} must not be placeholder text (got {value!r})")
+    return text
+
+
+def coerce_to_list(value: Any) -> Any:
+    """Wrap a lone scalar/dict into a one-element list; pass lists/None through.
+
+    Mirrors ``api.schemas.v1.do._coerce_to_list``: an agent that passes a single
+    string (or dict) where a list is declared is making the well-intentioned
+    single-item mistake — wrap it rather than reject it.
+    """
+    if value is None or isinstance(value, list):
+        return value
+    if isinstance(value, str | dict):
+        return [value]
+    return value
diff --git a/tests/unit/foundation/policy/content/test_content_models.py b/tests/unit/foundation/policy/content/test_content_models.py
new file mode 100644
index 00000000..0714fed3
--- /dev/null
+++ b/tests/unit/foundation/policy/content/test_content_models.py
@@ -0,0 +1,220 @@
+"""Validation tests for the structured content schema models."""
+
+from __future__ import annotations
+
+import pytest
+from roboco.foundation.policy.content import (
+    AuditorNote,
+    ContentValidationError,
+    DeveloperNote,
+    DocNote,
+    PrReviewContent,
+    QaNote,
+    ResumptionNote,
+    TaskDescription,
+    validate_content,
+)
+from roboco.foundation.policy.content.enums import Severity, Verdict
+
+# --------------------------------------------------------------------------- #
+# Valid construction
+# --------------------------------------------------------------------------- #
+
+
+def test_pr_review_valid() -> None:
+    c = PrReviewContent(
+        summary="The change is correct and covered by tests.",
+        findings=[
+            {
+                "file": "roboco/services/git.py",
+                "line": 42,
+                "severity": "major",
+                "expected": "raises on 422",
+                "actual": "swallows the error",
+            }
+        ],
+        verdict="changes_requested",
+    )
+    assert c.findings[0].severity is Severity.MAJOR
+    assert c.verdict is Verdict.CHANGES_REQUESTED
+
+
+def test_qa_valid() -> None:
+    c = QaNote(
+        summary="Reviewed all acceptance criteria against the diff.",
+        ac_verdicts=[
+            {"criterion": "AC1 returns 400", "status": "verified", "how": "test passes"}
+        ],
+        verdict="passed",
+    )
+    assert c.ac_verdicts[0].status == "verified"
+
+
+def test_task_description_valid() -> None:
+    c = TaskDescription(
+        objective="Add a structured PR-review comment format.",
+        what_this_builds=["a reviewer schema"],
+        the_work=[
+            {
+                "team": "backend",
+                "summary": "schema + gateway",
+                "items": ["model", "verb"],
+            }
+        ],
+        acceptance_criteria=["reviewer notes land in their own slot"],
+    )
+    assert c.the_work[0].team.value == "backend"
+
+
+def test_resumption_valid() -> None:
+    c = ResumptionNote(
+        done="schema landed", next="wire the gateway", where_to_look=["models.py"]
+    )
+    assert c.next == "wire the gateway"
+
+
+# --------------------------------------------------------------------------- #
+# Rejection
+# --------------------------------------------------------------------------- #
+
+
+def test_validate_content_unknown_type() -> None:
+    with pytest.raises(ContentValidationError) as exc:
+        validate_content("nonsense", {})
+    assert exc.value.field == "content_type"
+
+
+def test_pr_review_missing_summary_rejected() -> None:
+    with pytest.raises(ContentValidationError) as exc:
+        validate_content("pr_review", {"verdict": "approved"})
+    assert exc.value.field == "summary"
+
+
+def test_pr_review_trivial_summary_rejected() -> None:
+    with pytest.raises(ContentValidationError):
+        validate_content("pr_review", {"summary": "wip", "verdict": "approved"})
+
+
+def test_pr_review_negative_verdict_requires_findings() -> None:
+    with pytest.raises(ContentValidationError):
+        validate_content(
+            "pr_review",
+            {
+                "summary": "Looks broken but no detail given here.",
+                "verdict": "failed",
+                "findings": [],
+            },
+        )
+
+
+def test_pr_review_approved_allows_empty_findings() -> None:
+    c = validate_content(
+        "pr_review",
+        {"summary": "All criteria met, nothing to change.", "verdict": "approved"},
+    )
+    assert isinstance(c, PrReviewContent)
+
+
+def test_qa_verdict_must_be_pass_or_fail() -> None:
+    with pytest.raises(ContentValidationError) as exc:
+        validate_content(
+            "qa",
+            {
+                "summary": "Reviewed the acceptance criteria thoroughly.",
+                "ac_verdicts": [
+                    {"criterion": "c", "status": "verified", "how": "test"}
+                ],
+                "verdict": "approved",
+            },
+        )
+    assert exc.value.field == "verdict"
+
+
+def test_qa_requires_ac_verdicts() -> None:
+    with pytest.raises(ContentValidationError) as exc:
+        validate_content(
+            "qa",
+            {
+                "summary": "Reviewed everything carefully here.",
+                "ac_verdicts": [],
+                "verdict": "passed",
+            },
+        )
+    assert exc.value.field == "ac_verdicts"
+
+
+def test_task_description_requires_work() -> None:
+    with pytest.raises(ContentValidationError) as exc:
+        validate_content(
+            "task_description",
+            {
+                "objective": "Build the thing properly.",
+                "the_work": [],
+                "acceptance_criteria": ["x"],
+            },
+        )
+    assert exc.value.field == "the_work"
+
+
+def test_work_unit_rejects_non_cell_team() -> None:
+    with pytest.raises(ContentValidationError):
+        validate_content(
+            "task_description",
+            {
+                "objective": "Build the thing properly.",
+                "the_work": [{"team": "board", "summary": "do it", "items": ["a"]}],
+                "acceptance_criteria": ["x"],
+            },
+        )
+
+
+# --------------------------------------------------------------------------- #
+# Graceful coercion (lone scalar -> one-element list)
+# --------------------------------------------------------------------------- #
+
+
+def test_findings_single_dict_coerced_to_list() -> None:
+    c = validate_content(
+        "pr_review",
+        {
+            "summary": "One finding passed as a bare dict.",
+            "verdict": "changes_requested",
+            "findings": {
+                "file": "a.py",
+                "severity": "nit",
+                "expected": "trailing newline",
+                "actual": "no newline",
+            },
+        },
+    )
+    assert isinstance(c, PrReviewContent)
+    assert len(c.findings) == 1
+
+
+def test_where_to_look_single_string_coerced() -> None:
+    c = validate_content(
+        "resumption",
+        {
+            "done": "did the thing",
+            "next": "do next thing",
+            "where_to_look": "models.py",
+        },
+    )
+    assert isinstance(c, ResumptionNote)
+    assert c.where_to_look == ["models.py"]
+
+
+def test_developer_and_doc_and_auditor_models() -> None:
+    assert isinstance(
+        validate_content("developer", {"summary": "Implemented the schema layer."}),
+        DeveloperNote,
+    )
+    assert isinstance(
+        validate_content("doc", {"summary": "Documented the new endpoints."}), DocNote
+    )
+    assert isinstance(
+        validate_content(
+            "auditor", {"summary": "No concerns with this change.", "severity": "info"}
+        ),
+        AuditorNote,
+    )
diff --git a/tests/unit/foundation/policy/content/test_content_renderers.py b/tests/unit/foundation/policy/content/test_content_renderers.py
new file mode 100644
index 00000000..29798d98
--- /dev/null
+++ b/tests/unit/foundation/policy/content/test_content_renderers.py
@@ -0,0 +1,92 @@
+"""Markdown rendering tests for the structured content models.
+
+Assert structure (labeled sections + the content appears) rather than exact
+bytes, so the renderers stay refactorable.
+"""
+
+from __future__ import annotations
+
+from roboco.foundation.policy.content import (
+    PrReviewContent,
+    QaNote,
+    ResumptionNote,
+    TaskDescription,
+)
+
+
+def test_pr_review_renders_sections_and_findings_table() -> None:
+    md = PrReviewContent(
+        summary="The guard is missing on the 422 path.",
+        findings=[
+            {
+                "file": "roboco/services/git.py",
+                "line": 42,
+                "severity": "blocker",
+                "expected": "retry as COMMENT",
+                "actual": "raises",
+            }
+        ],
+        verdict="changes_requested",
+    ).render_markdown()
+    assert "## Summary" in md
+    assert "## Findings" in md
+    assert "| File | Line | Severity | Expected → Actual |" in md
+    assert "`roboco/services/git.py`" in md
+    assert "blocker" in md
+    assert "## Verdict" in md
+    assert "changes requested" in md
+
+
+def test_qa_renders_checklist() -> None:
+    md = QaNote(
+        summary="Verified every acceptance criterion against the diff.",
+        ac_verdicts=[
+            {
+                "criterion": "AC1 returns 400",
+                "status": "verified",
+                "how": "test passes",
+            },
+            {"criterion": "AC2 logs", "status": "failed", "how": "no log emitted"},
+        ],
+        verdict="failed",
+    ).render_markdown()
+    assert "## Acceptance Criteria" in md
+    assert "✅" in md and "❌" in md
+    assert "AC1 returns 400" in md
+    assert "## Verdict" in md
+
+
+def test_resumption_renders_done_next_where() -> None:
+    md = ResumptionNote(
+        done="schema module landed",
+        next="wire the gateway chokepoint",
+        where_to_look=[
+            "foundation/policy/content/models.py",
+            "services/content_notes.py",
+        ],
+    ).render_markdown()
+    assert "## Done" in md
+    assert "## Next" in md
+    assert "## Where to look" in md
+    assert "content_notes.py" in md
+    # No machine markers ever leak into the human resumption note.
+    assert "original_developer" not in md
+    assert "documenter:" not in md
+
+
+def test_task_description_renders_all_sections() -> None:
+    md = TaskDescription(
+        objective="Add the structured content standard.",
+        what_this_builds=["a schema layer", "a gateway gate"],
+        the_work=[
+            {"team": "backend", "summary": "models + verbs", "items": ["model", "verb"]}
+        ],
+        notes=["reuse the Team enum"],
+        acceptance_criteria=["reviewer notes are isolated"],
+    ).render_markdown()
+    assert "## Objective" in md
+    assert "## What This Builds" in md
+    assert "## The Work" in md
+    assert "**Backend**" in md
+    assert "## Notes" in md
+    assert "## Acceptance Criteria" in md