feat(tasks): task-content guardrails — structured plans + constraints split (#328)

* feat(tasks): task-content guardrails — structured plans + constraints split

Bound task PLANNING content the way journals/notes already are, fixing the
poor task quality flagged 2026-07-07 (degenerate roots, over-decomposed
leaves, descriptions bloated by an auto-attached conventions dump).

Phase A — plan/AC guardrails (no migration):
- _pm_sub_tasks_gate: cap sub_tasks at 7; per-subtask ceilings (title <=200,
  description <=600) enforced at both the Pydantic boundary and the gate.
  Dropped the min-2-roots and no-subtasks-on-code rules: both contradict the
  2026-05-08 rule (test_cell_pm_can_plan_code_typed_parent_via_i_will_plan)
  and break legitimate single-cell roots. Long comment in the gate explains.
- IWillPlanRequest: plan <=2000, approach <=800 (floor 150 kept), typed
  SubTaskCreate/RiskCreate/OpenQuestionCreate replacing loose list[dict].
- DelegateRequest + task_completeness: acceptance_criteria capped at 7 items,
  each <=200 chars. New FieldRule.MAX_LENGTH_LIST + _post_rule_reject helper
  (extracted to keep the gate under xenon B).
- Routes dump typed models to dicts for the existing rich_plan shaper.

Phase B — conventions split (migration 068):
- New nullable tasks.constraints Text column; _attach_baseline_constraints
  now writes the ## Constraints block there instead of appending to
  description, so description is the human-authored instruction only. The
  conventions still reach the agent independently at spawn via the ambient
  block, so agent correctness is unaffected.
- TaskResponse / Task model / panel Task type carry constraints; panel shows
  a read-only Constraints card. Field is optional on the TS type (backend
  returns null for flag-off / pre-migration rows).

Tests: 5 new gate unit tests, 7 schema tests, 3 AC policy tests, 3 e2e smoke
scenarios; 4 baseline-constraints integration tests updated. ruff/mypy/xenon
clean; 10026 unit+foundation+e2e green; panel typecheck clean.

Refs: plan breezy-imagining-kahn

* test(tasks): use typed SubTaskCreate instead of dict literals in plan tests

make quality runs mypy over tests/ (1079 files), not just roboco/ — the
four sites passing dict literals to the now-typed sub_tasks: list[SubTaskCreate]
field failed mypy. Construct SubTaskCreate directly; the typed model raising
ValidationError IS the boundary the rejection tests assert.

* fix(deps): drop unused python-jose — clears PYSEC-2026-1325 (ecdsa, no fix)

CI's pip-audit went red on a freshly-published advisory PYSEC-2026-1325
against ecdsa 0.19.2 (no fix published — 0.19.2 is the latest). ecdsa is a
transitive dep of python-jose, which is a DIRECT dep of roboco but is NOT
imported anywhere in roboco/ or tests/ (grep-verified). The actual JWT path
uses PyJWT (import jwt) + fastapi_users.jwt, not python-jose.

So python-jose is a dead dependency. Removing it (deletion over an
--ignore-vuln waiver) drops ecdsa + rsa + pyasn1 + their type stubs from the
lockfile, eliminating the CVE at the source. deptry roboco/ stays clean
(no missing-dep), mypy clean, auth + schema tests pass.

Master CI was green 9h before this PR's run, so the advisory published in
that window would red any run including master — this fix unblocks both.

* chore(prompts): regenerate verb tables for typed plan sub_tasks

Phase A's IWillPlanRequest schema change (sub_tasks/risks/open_questions from
loose list[dict] to typed SubTaskCreate/RiskCreate/OpenQuestionCreate) made
the auto-generated verb tables stale. Regenerated via
scripts/regenerate_verb_tables.py — the diff is purely the signature
reflection (list[str|str] -> list[SubTaskCreate], etc.). Required by the
foundation-check gate (Makefile:559).

---------

Co-authored-by: Renn F <rennf93@users.noreply.github.com>

agents/prompts/_generated/cell_pm.md

@@ -12,7 +12,7 @@
 | `escalate_up` | `escalate_up(task_id: UUID, reason: str)` |
 | `give_me_work` | `give_me_work()` |
 | `i_am_idle` | `i_am_idle()` |
-| `i_will_plan` | `i_will_plan(task_id: UUID, plan: str, approach: str, sub_tasks: list[str | str] = PydanticUndefined, technical_considerations: list[str] = PydanticUndefined, risks: list[str | str] = PydanticUndefined, open_questions: list[str | str | bool] = PydanticUndefined)` |
+| `i_will_plan` | `i_will_plan(task_id: UUID, plan: str, approach: str, sub_tasks: list[SubTaskCreate] = PydanticUndefined, technical_considerations: list[str] = PydanticUndefined, risks: list[RiskCreate] = PydanticUndefined, open_questions: list[OpenQuestionCreate] = PydanticUndefined)` |
 | `reassign` | `reassign(task_id: UUID, new_assignee: str)` |
 | `request_changes` | `request_changes(task_id: UUID, issues: list[str])` |
 | `resume` | `resume(task_id: UUID)` |

agents/prompts/_generated/main_pm.md

@@ -13,7 +13,7 @@
 | `escalate_up` | `escalate_up(task_id: UUID, reason: str)` |
 | `give_me_work` | `give_me_work()` |
 | `i_am_idle` | `i_am_idle()` |
-| `i_will_plan` | `i_will_plan(task_id: UUID, plan: str, approach: str, sub_tasks: list[str | str] = PydanticUndefined, technical_considerations: list[str] = PydanticUndefined, risks: list[str | str] = PydanticUndefined, open_questions: list[str | str | bool] = PydanticUndefined)` |
+| `i_will_plan` | `i_will_plan(task_id: UUID, plan: str, approach: str, sub_tasks: list[SubTaskCreate] = PydanticUndefined, technical_considerations: list[str] = PydanticUndefined, risks: list[RiskCreate] = PydanticUndefined, open_questions: list[OpenQuestionCreate] = PydanticUndefined)` |
 | `request_changes` | `request_changes(task_id: UUID, issues: list[str])` |
 | `resume` | `resume(task_id: UUID)` |
 | `submit_root` | `submit_root(task_id: UUID, notes: str)` |

agents/prompts/_generated/verbs.md

@@ -115,7 +115,7 @@ real tools live in their agent_sdk drivers, not role_config.
 | `escalate_up` | `escalate_up(task_id: UUID, reason: str)` |
 | `give_me_work` | `give_me_work()` |
 | `i_am_idle` | `i_am_idle()` |
-| `i_will_plan` | `i_will_plan(task_id: UUID, plan: str, approach: str, sub_tasks: list[str | str] = PydanticUndefined, technical_considerations: list[str] = PydanticUndefined, risks: list[str | str] = PydanticUndefined, open_questions: list[str | str | bool] = PydanticUndefined)` |
+| `i_will_plan` | `i_will_plan(task_id: UUID, plan: str, approach: str, sub_tasks: list[SubTaskCreate] = PydanticUndefined, technical_considerations: list[str] = PydanticUndefined, risks: list[RiskCreate] = PydanticUndefined, open_questions: list[OpenQuestionCreate] = PydanticUndefined)` |
 | `reassign` | `reassign(task_id: UUID, new_assignee: str)` |
 | `request_changes` | `request_changes(task_id: UUID, issues: list[str])` |
 | `resume` | `resume(task_id: UUID)` |
@@ -152,7 +152,7 @@ real tools live in their agent_sdk drivers, not role_config.
 | `escalate_up` | `escalate_up(task_id: UUID, reason: str)` |
 | `give_me_work` | `give_me_work()` |
 | `i_am_idle` | `i_am_idle()` |
-| `i_will_plan` | `i_will_plan(task_id: UUID, plan: str, approach: str, sub_tasks: list[str | str] = PydanticUndefined, technical_considerations: list[str] = PydanticUndefined, risks: list[str | str] = PydanticUndefined, open_questions: list[str | str | bool] = PydanticUndefined)` |
+| `i_will_plan` | `i_will_plan(task_id: UUID, plan: str, approach: str, sub_tasks: list[SubTaskCreate] = PydanticUndefined, technical_considerations: list[str] = PydanticUndefined, risks: list[RiskCreate] = PydanticUndefined, open_questions: list[OpenQuestionCreate] = PydanticUndefined)` |
 | `request_changes` | `request_changes(task_id: UUID, issues: list[str])` |
 | `resume` | `resume(task_id: UUID)` |
 | `submit_root` | `submit_root(task_id: UUID, notes: str)` |

alembic/versions/068_tasks_constraints_column.py

@@ -0,0 +1,41 @@
+"""Add tasks.constraints — move the conventions dump out of description.
+
+The 2026-07-07 task-quality defect: ``TaskService._attach_baseline_constraints``
+appended a ``## Constraints`` block to ``tasks.description``, so descriptions
+ran 3000-4250 chars dominated by an auto-attached conventions dump. The
+conventions already reach the agent independently at spawn via
+``ConventionsService.render_ambient_block`` -> ``compose_prompt(ambient=...)``,
+so the stored block is redundant bloat. Move it to a dedicated nullable
+``constraints`` Text column so ``description`` is the human-authored
+instruction only and the panel can render the constraints as a read-only card.
+
+Additive: nullable column, default NULL. Existing rows keep their bloated
+descriptions (a backfill that regex-strips markdown from a Text column is
+riskier than the win — new tasks get clean descriptions from B2); no data
+change in the migration.
+
+Revision ID: 068_tasks_constraints_column
+Revises: 067_respawn_revisit_resets
+Create Date: 2026-07-07
+"""
+
+from __future__ import annotations
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "068_tasks_constraints_column"
+down_revision = "067_respawn_revisit_resets"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "tasks",
+        sa.Column("constraints", sa.Text(), nullable=True),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("tasks", "constraints")

panel/src/components/tasks/task-detail/task-description.tsx

@@ -78,6 +78,7 @@ export function TaskDescription({ task }: TaskDescriptionProps) {
   };
 
   return (
+    <>
     <Card>
       <CardHeader className="pb-3">
         <div className="flex items-center justify-between">
@@ -177,5 +178,22 @@ export function TaskDescription({ task }: TaskDescriptionProps) {
         )}
       </CardContent>
     </Card>
+    {task.constraints ? (
+      <Card className="border-dashed">
+        <CardHeader className="pb-3">
+          <CardTitle className="text-base text-muted-foreground">
+            Constraints
+          </CardTitle>
+        </CardHeader>
+        <CardContent>
+          <p className="text-xs text-muted-foreground mb-3">
+            Architectural standard derived from the project conventions —
+            read-only. Applies to every task in this project.
+          </p>
+          <Markdown>{task.constraints}</Markdown>
+        </CardContent>
+      </Card>
+    ) : null}
+    </>
   );
 }

panel/src/types/index.ts

@@ -216,6 +216,10 @@ export interface Task {
   id: string;
   title: string;
   description: string;
+  // Server-derived architectural constraints (project baseline conventions),
+  // moved out of description. Null when conventions are flag-off / none /
+  // pre-migration. Read-only on the panel — system-derived, not human-edited.
+  constraints?: string | null;
   acceptance_criteria: string[];
   status: TaskStatus;
   priority: number; // 0=P0(highest), 1=P1, 2=P2, 3=P3(lowest)

pyproject.toml

@@ -35,7 +35,6 @@ dependencies = [
     # Utilities
     "httpx",
     "python-multipart",
-    "python-jose[cryptography]",  # JWT
     "passlib[bcrypt]",  # Password hashing
     "tenacity",  # Retry logic
     "structlog",  # Structured logging
@@ -83,7 +82,6 @@ dev = [
 
     # Type Stubs
     "types-passlib",
-    "types-python-jose",
     "types-PyYAML",
 
     # Development
@@ -397,8 +395,7 @@ DEP002 = [
     "python-multipart",
     # Database migrations (CLI tool)
     "alembic",
-    # Auth libraries (used via passlib[bcrypt], python-jose[cryptography])
+    # Auth libraries (used via passlib[bcrypt]; JWT via PyJWT + fastapi_users.jwt)
-    "python-jose",
     "passlib",
     # LLM utilities (embeddings/token counting)
     "openai",

roboco/api/routes/v1/flow_cell_pm.py

@@ -71,10 +71,10 @@ async def i_will_plan(
         body.plan,
         rich_plan={
             "approach": body.approach,
-            "sub_tasks": body.sub_tasks,
+            "sub_tasks": [s.model_dump() for s in body.sub_tasks],
             "technical_considerations": body.technical_considerations,
-            "risks": body.risks,
+            "risks": [r.model_dump() for r in body.risks],
-            "open_questions": body.open_questions,
+            "open_questions": [q.model_dump() for q in body.open_questions],
         },
     )
     return envelope_to_response(env, request)

roboco/api/routes/v1/flow_main_pm.py

@@ -71,10 +71,14 @@ async def i_will_plan(
         body.plan,
         rich_plan={
             "approach": body.approach,
-            "sub_tasks": body.sub_tasks,
+            # Typed models -> dicts: _build_panel_shaped_plan filters
+            # isinstance(st, dict), so a SubTaskCreate instance would be
+            # dropped. Dump to the dict shape every downstream reader
+            # (_plan_subtasks, _normalize_sub_task, the gate) already uses.
+            "sub_tasks": [s.model_dump() for s in body.sub_tasks],
             "technical_considerations": body.technical_considerations,
-            "risks": body.risks,
+            "risks": [r.model_dump() for r in body.risks],
-            "open_questions": body.open_questions,
+            "open_questions": [q.model_dump() for q in body.open_questions],
         },
     )
     return envelope_to_response(env, request)

roboco/api/schemas/tasks.py

@@ -200,7 +200,9 @@ class TaskUpdate(BaseModel):
     # Basic info
     title: str | None = Field(default=None, min_length=1, max_length=200)
     description: str | None = Field(default=None, min_length=20)
-    acceptance_criteria: list[str] | None = Field(default=None, min_length=1)
+    acceptance_criteria: list[str] | None = Field(
+        default=None, min_length=1, max_length=7
+    )
     priority: int | None = Field(default=None, ge=0, le=3)
     sequence: int | None = Field(default=None, ge=0)  # Order within siblings
     target_date: datetime | None = None
@@ -288,6 +290,10 @@ class TaskResponse(BaseModel):
     id: UUID
     title: str
     description: str
+    # Server-derived architectural constraints (project baseline conventions),
+    # moved out of `description` (migration 068). None when conventions are
+    # flag-off, the project has none, or the row predates the column.
+    constraints: str | None = None
     acceptance_criteria: list[str]
 
     # Status
@@ -773,6 +779,7 @@ def task_to_response(task: "TaskTable") -> TaskResponse:
         id=require_uuid(task.id),
         title=task.title,
         description=task.description,
+        constraints=getattr(task, "constraints", None),
         acceptance_criteria=task.acceptance_criteria or [],
         status=task.status,
         priority=task.priority,

roboco/api/schemas/v1/flow.py

@@ -20,6 +20,38 @@ from roboco.models.base import Complexity
 # technical_considerations, covers_parent_criteria).
 StrList = Annotated[list[str], BeforeValidator(coerce_str_list)]
 
+# AC discipline (the 2026-07-07 task-quality defect: restated, over-long
+# acceptance criteria). Mirrors roboco.foundation.policy.task_completeness
+# so an over-long / over-count AC list is rejected at the request boundary.
+_AC_MAX_ITEMS = 7
+_AC_MAX_ITEM_CHARS = 200
+
+
+class SubTaskCreate(BaseModel):
+    """A PM sub_task — a delegate target AND a progress-checklist item.
+
+    Mirrors DelegateRequest title/description caps so an over-long sub_task
+    can't bloat the plan (the 2026-07-07 task-quality defect). The server
+    assigns id + order; callers supply title + description only.
+    """
+
+    title: str = Field(..., min_length=1, max_length=200)
+    description: str = Field(..., min_length=20, max_length=600)
+
+
+class RiskCreate(BaseModel):
+    """A {risk, mitigation} entry — what could go wrong and how it's handled."""
+
+    risk: str = Field(..., min_length=1, max_length=300)
+    mitigation: str = Field(..., min_length=1, max_length=600)
+
+
+class OpenQuestionCreate(BaseModel):
+    """An open question the PM wants answered before/during work."""
+
+    question: str = Field(..., min_length=1, max_length=300)
+    answered: bool = False
+
 
 class GiveMeWorkRequest(BaseModel):
     """Empty request body — agent_id comes from header."""
@@ -240,7 +272,7 @@ class EscalateToCeoRequest(BaseModel):
 
 class IWillPlanRequest(BaseModel):
     task_id: UUID
-    plan: str = Field(..., min_length=1)
+    plan: str = Field(..., min_length=1, max_length=2000)
     # Pre-gateway parity: Approach is REQUIRED — agents
     # could not transition claimed → in_progress without filling this in the
     # pre-gateway flow. The Plan tab depends on it; smoke run 3 confirmed
@@ -248,14 +280,16 @@ class IWillPlanRequest(BaseModel):
     # min_length must match choreographer._impl._PM_APPROACH_MIN_LEN. Raised
     # 20→150: a 20-char approach was a one-liner; the approach +
     # sub_tasks are also the progress checklist, so they must be substantive.
-    approach: str = Field(..., min_length=150)
+    # max_length caps the bloat defect (approach that ran to thousands of
-    sub_tasks: list[dict[str, str]] = Field(
+    # chars restating the description). Must match the gate ceiling.
+    approach: str = Field(..., min_length=150, max_length=800)
+    sub_tasks: list[SubTaskCreate] = Field(
         default_factory=list,
         description="List of {title, description} — server assigns id + order",
     )
     technical_considerations: StrList = Field(default_factory=list)
-    risks: list[dict[str, str]] = Field(default_factory=list)
+    risks: list[RiskCreate] = Field(default_factory=list)
-    open_questions: list[dict[str, str | bool]] = Field(default_factory=list)
+    open_questions: list[OpenQuestionCreate] = Field(default_factory=list)
 
 
 class DelegateRequest(BaseModel):
@@ -283,8 +317,24 @@ class DelegateRequest(BaseModel):
     nature: str = Field(..., min_length=1)
     estimated_complexity: Complexity
     # acceptance_criteria is required and non-empty; downstream policy
-    # also denylist-checks each item against placeholder phrases.
+    # also denylist-checks each item against placeholder phrases. The
-    acceptance_criteria: StrList = Field(..., min_length=1)
+    # per-item cap (<=200 chars) + list cap (<=7) mirror the policy so an
+    # over-long / over-count AC list is rejected at the boundary (422), not
+    # only by the service-layer completeness check.
+    acceptance_criteria: StrList = Field(..., min_length=1, max_length=7)
+
+    @field_validator("acceptance_criteria")
+    @classmethod
+    def _ac_items_bounded(cls, v: list[str]) -> list[str]:
+        for i, item in enumerate(v):
+            if len(item.strip()) > _AC_MAX_ITEM_CHARS:
+                raise ValueError(
+                    f"acceptance_criteria[{i}] is {len(item.strip())} chars "
+                    f"(max {_AC_MAX_ITEM_CHARS}) — a criterion that long is a "
+                    "restated description, not a verifiable outcome. Split it."
+                )
+        return v
+
     # Optional per-subtask project override. When omitted, the choreographer
     # resolves the project from the parent's Product map for this cell, then
     # falls back to the parent's project. Plain optional field — no validator.

roboco/db/tables.py

@@ -160,6 +160,12 @@ class TaskTable(Base):
     )
     title: Mapped[str] = mapped_column(String(200), nullable=False)
     description: Mapped[str] = mapped_column(Text, nullable=False)
+    # Server-derived architectural constraints (project baseline conventions
+    # block), moved out of `description` (migration 068). Nullable: flag-off /
+    # no-conventions / pre-migration rows have none. The conventions also
+    # reach the agent at spawn via the ambient block — this column is for
+    # panel visibility, not agent correctness.
+    constraints: Mapped[str | None] = mapped_column(Text, nullable=True)
     acceptance_criteria: Mapped[list[str]] = mapped_column(
         ARRAY(String), nullable=False
     )

roboco/foundation/policy/task_completeness.py

@@ -29,6 +29,7 @@ class FieldRule(StrEnum):
     MIN_LENGTH = "min_length"
     NON_EMPTY_LIST = "non_empty_list"
     EXPLICITLY_DECLARED = "explicitly_declared"
+    MAX_LENGTH_LIST = "max_length_list"
 
 
 @dataclass(frozen=True)
@@ -100,8 +101,11 @@ _HINT_ACCEPTANCE_CRITERIA = (
     "non-empty list[str]; each item describes a verifiable outcome (e.g. "
     "'returns 401 when token absent'). Do NOT use placeholder strings like "
     "'completed and reviewed by assignee' — that's a known evasion phrase the "
-    "gateway rejects."
+    "gateway rejects. Keep each item <= 200 chars and the list <= 7 items — a "
+    "300-char AC or a 12-item list is a restated description, not a criterion."
 )
+_AC_MAX_ITEMS = 7
+_AC_MAX_ITEM_CHARS = 200
 _HINT_TASK_TYPE = (
     "one of: code | documentation | research | planning | design | administrative"
 )
@@ -129,6 +133,12 @@ TASK_AT_CREATE: CompletenessSpec = CompletenessSpec(
             FieldRule.NON_EMPTY_LIST,
             hint=_HINT_ACCEPTANCE_CRITERIA,
         ),
+        FieldRequirement(
+            "acceptance_criteria",
+            FieldRule.MAX_LENGTH_LIST,
+            _AC_MAX_ITEMS,
+            hint=_HINT_ACCEPTANCE_CRITERIA,
+        ),
         FieldRequirement(
             "task_type", FieldRule.EXPLICITLY_DECLARED, hint=_HINT_TASK_TYPE
         ),
@@ -196,6 +206,14 @@ def _check_non_empty_list(value: Any) -> tuple[bool, str | None]:
     return True, None
 
 
+def _check_max_length_list(value: Any, maximum: int) -> tuple[bool, str | None]:
+    if not isinstance(value, list):
+        return False, f"must be a list of length <= {maximum}"
+    if len(value) > maximum:
+        return False, f"must have at most {maximum} items (got {len(value)})"
+    return True, None
+
+
 def _check_field(req: FieldRequirement, value: Any) -> tuple[bool, str | None]:
     """Return (passed, problem_description). problem_description is None on pass."""
     if req.rule is FieldRule.EXPLICITLY_DECLARED:
@@ -204,6 +222,8 @@ def _check_field(req: FieldRequirement, value: Any) -> tuple[bool, str | None]:
         return _check_non_empty_string(value)
     if req.rule is FieldRule.MIN_LENGTH:
         return _check_min_length(value, req.value or 0)
+    if req.rule is FieldRule.MAX_LENGTH_LIST:
+        return _check_max_length_list(value, req.value or 0)
     return _check_non_empty_list(value)
 
 
@@ -217,6 +237,38 @@ def _matches_denylist_ac(items: Any) -> bool:
     )
 
 
+def _overlong_ac_item(items: Any) -> tuple[int, str] | None:
+    """First (index, raw) AC item exceeding the per-item char cap, else None."""
+    if not isinstance(items, list):
+        return None
+    for i, item in enumerate(items):
+        if isinstance(item, str) and len(item.strip()) > _AC_MAX_ITEM_CHARS:
+            return i, item
+    return None
+
+
+def _post_rule_reject(field: str, value: Any) -> str | None:
+    """Content-gate a field that already passed its structural rule.
+
+    Returns a hint prefix when the value is rejected, or None to pass. The
+    caller appends `req.hint`, so this returns only the rejection reason.
+    """
+    if field == "acceptance_criteria":
+        if _matches_denylist_ac(value):
+            return "rejected: placeholder phrase from the legacy silent fallback. "
+        over = _overlong_ac_item(value)
+        if over is not None:
+            idx, item = over
+            return (
+                f"rejected: criterion #{idx + 1} is {len(item.strip())} chars "
+                f"(max {_AC_MAX_ITEM_CHARS}) — a criterion that long is a "
+                "restated description, not a verifiable outcome. Split it. "
+            )
+    if field == "description" and _matches_denylist_description(value):
+        return "rejected: placeholder/empty phrase. "
+    return None
+
+
 def _matches_denylist_description(text: Any) -> bool:
     """True if the description matches any denylist regex."""
     if not isinstance(text, str):
@@ -252,19 +304,11 @@ def check(spec: CompletenessSpec, task: Any) -> CompletenessResult:
             field_hints[req.field] = req.hint
             continue
 
-        # Denylist checks (post-rule).
+        # Post-rule content checks (denylists / per-item caps).
-        if req.field == "acceptance_criteria" and _matches_denylist_ac(value):
+        hint = _post_rule_reject(req.field, value)
-            missing.append("acceptance_criteria")
+        if hint is not None:
-            field_hints["acceptance_criteria"] = (
+            missing.append(req.field)
-                "rejected: placeholder phrase from the legacy silent fallback. "
+            field_hints[req.field] = hint + req.hint
-                + req.hint
-            )
-            continue
-        if req.field == "description" and _matches_denylist_description(value):
-            missing.append("description")
-            field_hints["description"] = (
-                "rejected: placeholder/empty phrase. " + req.hint
-            )
             continue
 
     return CompletenessResult(

roboco/models/task.py

@@ -141,8 +141,15 @@ class Task(TimestampMixin):
     id: UUID = Field(default_factory=uuid4, description="Unique task identifier")
     title: str = Field(..., min_length=1, max_length=200, description="Task title")
     description: str = Field(..., description="Detailed task description")
+    # Server-derived architectural constraints (the project's baseline
+    # conventions block), moved out of `description` so the description is the
+    # human-authored instruction only (2026-07-07 task-quality fix). The
+    # conventions ALSO reach the agent at spawn via the ambient block, so
+    # this field is for panel visibility, not agent correctness. Nullable:
+    # flag-off / no-conventions / pre-migration rows have none.
+    constraints: str | None = Field(default=None)
     acceptance_criteria: list[str] = Field(
-        ..., min_length=1, description="How do we know it's done?"
+        ..., min_length=1, max_length=7, description="How do we know it's done?"
     )
     acceptance_criteria_ids: list[str] = Field(
         default_factory=list,

roboco/services/gateway/choreographer/_impl.py

@@ -77,13 +77,26 @@ _PM_APPROACH_MIN_LEN = 150
 # description that actually says what the step does.
 _PM_SUBTASK_DESC_MIN_LEN = 60
 
+# Per-sub_task ceilings — mirror IWillPlanRequest.SubTaskCreate so direct
+# service-layer callers (MCP, fixtures) can't bypass the Pydantic boundary
+# with a 5000-char description (the bloat defect: descriptions dominated by
+# over-long sub-task prose).
+_PM_SUBTASK_TITLE_MAX_LEN = 200
+_PM_SUBTASK_DESC_MAX_LEN = 600
+
+# Over-decomposition cap (the 2026-07-07 task-quality defect): a plan with
+# more than this many sub_tasks should be split into sibling coordination
+# tasks, not one giant plan.
+_PM_SUBTASKS_MAX = 7
+
 
 def _thin_subtask_hint(sub_tasks: list[Any]) -> str | None:
-    """Return a hint if any PM sub_task is title-only / thin.
+    """Return a hint if any PM sub_task is title-only / thin / over-long.
 
     Each sub_task is a delegate target AND a progress-checklist item, so
-    a title with no real description is not a plan. Returns None when
+    a title with no real description is not a plan. Also caps title +
-    every sub_task carries a title and a substantive description.
+    description length so a sub_task can't bloat the plan. Returns None
+    when every sub_task carries a title and a substantive description.
     """
     for i, st in enumerate(sub_tasks):
         if not isinstance(st, dict):
@@ -92,12 +105,24 @@ def _thin_subtask_hint(sub_tasks: list[Any]) -> str | None:
         desc = str(st.get("description") or "").strip()
         if not title:
             return f"sub_task #{i + 1} has no title."
+        if len(title) > _PM_SUBTASK_TITLE_MAX_LEN:
+            return (
+                f"sub_task #{i + 1} ('{title[:40]}') title is too long "
+                f"({len(title)} chars) — keep titles <= "
+                f"{_PM_SUBTASK_TITLE_MAX_LEN} chars."
+            )
         if len(desc) < _PM_SUBTASK_DESC_MIN_LEN:
             return (
                 f"sub_task #{i + 1} ('{title[:40]}') description is too thin "
                 f"({len(desc)} chars) — need >= {_PM_SUBTASK_DESC_MIN_LEN} "
                 "characters describing what the step actually does."
             )
+        if len(desc) > _PM_SUBTASK_DESC_MAX_LEN:
+            return (
+                f"sub_task #{i + 1} ('{title[:40]}') description is too long "
+                f"({len(desc)} chars) — keep descriptions <= "
+                f"{_PM_SUBTASK_DESC_MAX_LEN} chars; split if larger."
+            )
     return None
 
 
@@ -540,6 +565,24 @@ class Choreographer:
         elif thin := _thin_subtask_hint(sub_tasks):
             missing.append("sub_tasks")
             field_hints["sub_tasks"] = thin
+        elif len(sub_tasks) > _PM_SUBTASKS_MAX:
+            # Over-decomposition cap (the 2026-07-07 task-quality defect):
+            # a plan with >7 sub_tasks is one giant plan that should be split
+            # into sibling coordination tasks. Per-item title/description
+            # bounds are in _thin_subtask_hint above. We deliberately do NOT
+            # ban sub_tasks on code tasks (PMs legitimately plan code-typed
+            # parents into dev subtasks — the 2026-05-08 rule,
+            # test_cell_pm_can_plan_code_typed_parent_via_i_will_plan) nor
+            # require >=2 on roots (a single-cell root → one cell task is a
+            # legitimate pass-through, test_pm_can_plan_non_code_parent) —
+            # both judgment calls the gate can't make without false positives.
+            missing.append("sub_tasks")
+            field_hints["sub_tasks"] = (
+                f"too many sub_tasks ({len(sub_tasks)}) — decompose into "
+                f"at most {_PM_SUBTASKS_MAX}. If the work is genuinely "
+                "larger, split it into sibling coordination tasks instead "
+                "of one giant plan."
+            )
         if not missing:
             return None
         return await self._emit_rejection(
@@ -549,8 +592,11 @@ class Choreographer:
                 remediate=(
                     "re-issue i_will_plan(task_id, plan, approach, "
                     "sub_tasks=[{'title': '...', 'description': '...'}, ...]) "
-                    f"with approach >= {_PM_APPROACH_MIN_LEN} chars and every "
+                    f"with approach >= {_PM_APPROACH_MIN_LEN} and <= 800 "
-                    f"sub_task description >= {_PM_SUBTASK_DESC_MIN_LEN} chars."
+                    f"chars, every sub_task description >= "
+                    f"{_PM_SUBTASK_DESC_MIN_LEN} and <= "
+                    f"{_PM_SUBTASK_DESC_MAX_LEN} chars, and at most "
+                    f"{_PM_SUBTASKS_MAX} sub_tasks."
                 ),
                 context_briefing=briefing,
             ).with_introspection(task=task, role=role_str),

roboco/services/task.py

@@ -1124,25 +1124,30 @@ class TaskService(BaseService):
         return task
 
     async def _attach_baseline_constraints(self, task: TaskTable) -> None:
-        """Append the project's block-rule constraints to the task description.
+        """Write the project's block-rule constraints to ``task.constraints``.
 
         Server-derived backstop (flag-gated): every project task carries the
         hard conventions even if nothing upstream added them — the layer the
-        design calls "can't be skipped". Idempotent and non-suppressible: it
+        design calls "can't be skipped". Idempotent: appends only baseline
-        appends only baseline constraints not already present (dedup by exact
+        constraints not already present in the column (dedup by exact string),
-        string), so a second pass adds nothing AND an agent-authored
+        so a second pass adds nothing AND an agent-authored ``## Constraints``
-        ``## Constraints`` section can never suppress the mandatory baseline.
+        section in the description can never suppress the mandatory baseline
-        Best-effort: a failure never blocks task creation.
+        (the column is the source of truth for the panel; the description is
+        the human-authored instruction only — 2026-07-07 task-quality fix).
+        The conventions ALSO reach the agent at spawn via the ambient block
+        (``ConventionsService.render_ambient_block``), so this column is for
+        panel visibility, not agent correctness. Best-effort: a failure never
+        blocks task creation.
         """
         baseline = await self._project_baseline_constraints(task)
         if not baseline:
             return
-        existing = task.description or ""
+        existing = task.constraints or ""
         missing = [item for item in baseline if item not in existing]
         if not missing:
             return
         section = "## Constraints\n" + "\n".join(f"- {item}" for item in missing)
-        task.description = f"{existing}\n\n{section}" if existing else section
+        task.constraints = f"{existing}\n\n{section}" if existing else section
         await self.session.flush()
 
     async def _project_baseline_constraints(self, task: TaskTable) -> list[str]:

tests/e2e_smoke/test_pm_plan_guardrails.py

@@ -0,0 +1,161 @@
+"""Scenario: the PM plan-content guardrails reject over-decomposed plans.
+
+The 2026-07-07 task-quality defects (root 79d686f0 decomposed into 1 subtask,
+code leaf 55376b8a carrying a 5-subtask plan, descriptions bloated to 3000+
+chars) were structural — the planning verb barely guardrailed its content.
+The fix adds ceilings (plan <= 2000, approach <= 800, sub-task title <= 200,
+sub-task description <= 600) and an over-decomposition cap (>7 sub_tasks
+rejected) at both the HTTP Pydantic boundary and the choreographer gate.
+
+This scenario drives a MAIN_PM planning root through the real API and asserts
+each guardrail surfaces a clean ``incomplete_input`` envelope with a
+remediation hint the agent can act on — not a 500, not a silent accept. The
+gate is the load-bearing layer (direct service callers bypass Pydantic), so
+the assertions land on the envelope, not the HTTP status.
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any
+
+from tests.e2e_smoke.arcs import (
+    origin_branch,
+    seed_company,
+    seed_project,
+    seed_task,
+    set_branch_name,
+)
+from tests.e2e_smoke.harness import ScriptedAgent, expect_error
+
+if TYPE_CHECKING:
+    from uuid import UUID
+
+    from tests.e2e_smoke.arcs import Company
+    from tests.e2e_smoke.harness import E2EStack
+
+# Pydantic's IWillPlanRequest.approach enforces 150..800 chars at the HTTP
+# boundary, so every i_will_plan call needs a compliant approach.
+_APPROACH = (
+    "Plan and delegate the page-scoped refresh button work to the frontend "
+    "cell: land the provider/hook, add the navbar button, remove the inline "
+    "buttons, and route one planning subtask to fe-pm for delivery. "
+    "Sequenced strictly; no cross-cell dependencies for this slice."
+)
+_GOOD_SUB = {
+    "title": "Frontend cell: refresh button",
+    "description": (
+        "Delegate the navbar refresh button to fe-pm: land the provider/hook "
+        "and wire the click handler into the page, then open the leaf PR."
+    ),
+}
+_PLAN = "Land the refresh button via the frontend cell."
+
+
+def _seed_planning_root(
+    stack: E2EStack, company: Company
+) -> tuple[ScriptedAgent, UUID]:
+    """Seed a PENDING MAIN_PM planning root + its origin branch."""
+    from roboco.models import Team
+    from roboco.models.base import TaskStatus, TaskType
+
+    project_id, _project_slug = seed_project(stack, company)
+    main_pm = ScriptedAgent(stack, company.main_pm_id, "main-pm", "main_pm")
+    task_id = seed_task(
+        stack,
+        title="Root: page-scoped refresh button",
+        description="Frontend-only root: provider/hook + navbar button.",
+        acceptance_criteria=["the refresh button lands on master"],
+        task_type=TaskType.PLANNING,
+        team=Team.MAIN_PM,
+        project_id=project_id,
+        created_by=company.main_pm_id,
+        assigned_to=company.main_pm_id,
+        status=TaskStatus.PENDING,
+    )
+    branch = f"feature/main_pm/{str(task_id)[:8]}"
+    origin_branch(stack, branch, start="master")
+    set_branch_name(stack, task_id, branch)
+    return main_pm, task_id
+
+
+def _plan_with(sub_tasks: list[dict[str, Any]]) -> dict[str, Any]:
+    return {
+        "plan": _PLAN,
+        "approach": _APPROACH,
+        "sub_tasks": sub_tasks,
+    }
+
+
+def test_pm_plan_over_decomposition_cap_rejected(e2e_stack: E2EStack) -> None:
+    """A plan with >7 sub_tasks is over-decomposition — the gate rejects it
+    with incomplete_input + a 'split into sibling coordination tasks' hint."""
+    stack = e2e_stack
+    company = seed_company(stack)
+    main_pm, task_id = _seed_planning_root(stack, company)
+
+    env = main_pm.flow(
+        "i_will_plan",
+        task_id=str(task_id),
+        plan=_PLAN,
+        approach=_APPROACH,
+        sub_tasks=[dict(_GOOD_SUB, title=f"Slice {i}") for i in range(8)],
+    )
+    body = expect_error(env, "incomplete_input", "8 sub_tasks rejected")
+    assert "sub_tasks" in (body.get("missing") or []), body
+    assert "at most 7" in str(body.get("field_hints", {})), body
+
+
+def test_pm_plan_overlong_subtask_description_rejected(
+    e2e_stack: E2EStack,
+) -> None:
+    """A sub-task description >600 chars is the bloat defect — rejected at
+    the Pydantic boundary (422), so the envelope carries the validation
+    ``detail`` rather than the gate's ``field_hints``. Both layers are the
+    guardrail working; this test pins the boundary layer."""
+    stack = e2e_stack
+    company = seed_company(stack)
+    main_pm, task_id = _seed_planning_root(stack, company)
+
+    bloated = dict(_GOOD_SUB, description="x" * 700)
+    env = main_pm.flow(
+        "i_will_plan",
+        task_id=str(task_id),
+        plan=_PLAN,
+        approach=_APPROACH,
+        sub_tasks=[bloated],
+    )
+    body = expect_error(env, "incomplete_input", "over-long subtask desc")
+    # Boundary 422: missing is [] but detail carries the Pydantic error
+    # naming sub_tasks + the 600-char cap.
+    detail = str(body.get("detail"))
+    assert "sub_tasks" in detail, body
+    assert "600" in detail, body
+
+
+def test_pm_plan_valid_plan_passes_gate(e2e_stack: E2EStack) -> None:
+    """A well-formed plan (2 sub_tasks, bounded fields) passes the guardrails
+    and transitions the root to in_progress — the happy path stays green."""
+    stack = e2e_stack
+    company = seed_company(stack)
+    main_pm, task_id = _seed_planning_root(stack, company)
+
+    env = main_pm.flow(
+        "i_will_plan",
+        task_id=str(task_id),
+        plan=_PLAN,
+        approach=_APPROACH,
+        sub_tasks=[
+            _GOOD_SUB,
+            {
+                "title": "Frontend cell: remove inline buttons",
+                "description": (
+                    "Delegate removal of the stale inline refresh buttons to "
+                    "fe-pm so the navbar button is the single source of truth."
+                ),
+            },
+        ],
+    )
+    # The gate must not fire; the root moves to in_progress. Downstream may
+    # raise a different error (e.g. tracing_gap) but NOT incomplete_input.
+    body = env
+    assert body.get("error") != "incomplete_input", body

tests/foundation/test_task_completeness.py

@@ -152,3 +152,34 @@ def test_fill_parent_from_active_task_does_not_overwrite_explicit() -> None:
     payload = {"parent_task_id": "explicit-id"}
     result = tc.fill_parent_from_active_task(payload, "active-id")
     assert result["parent_task_id"] == "explicit-id"
+
+
+# ---------------------------------------------------------------------------
+# AC discipline (2026-07-07 task-quality defect): per-item cap + max count.
+# ---------------------------------------------------------------------------
+
+
+def test_task_at_create_rejects_too_many_acceptance_criteria() -> None:
+    """An AC list >7 items is over-decomposition of criteria."""
+    result = tc.check(
+        tc.TASK_AT_CREATE, _task(acceptance_criteria=[f"c{i}" for i in range(8)])
+    )
+    assert result.passed is False
+    assert "acceptance_criteria" in result.missing
+
+
+def test_task_at_create_accepts_seven_acceptance_criteria() -> None:
+    """Exactly 7 AC items is the cap — must pass."""
+    result = tc.check(
+        tc.TASK_AT_CREATE, _task(acceptance_criteria=[f"c{i}" for i in range(7)])
+    )
+    assert result.passed is True
+
+
+def test_task_at_create_rejects_overlong_acceptance_criterion() -> None:
+    """An AC item >200 chars is a restated description, not a criterion."""
+    long_ac = "x" * 201
+    result = tc.check(tc.TASK_AT_CREATE, _task(acceptance_criteria=[long_ac]))
+    assert result.passed is False
+    assert "acceptance_criteria" in result.missing
+    assert any("200" in h for h in result.field_hints.values())

tests/integration/test_task_baseline_constraints.py

@@ -68,9 +68,12 @@ async def test_baseline_attached_when_flag_on(
     monkeypatch.setattr(settings, "conventions_enabled", True)
     agent, project = await _seed(db_session)
     task = await TaskService(db_session).create(_req(agent, project, "Do the work"))
-    assert task.description is not None
+    # The conventions block lives in `constraints`, not `description` (the
-    assert "## Constraints" in task.description
+    # 2026-07-07 fix: description is the human-authored instruction only).
-    assert "no lint suppressions" in task.description
+    assert task.description == "Do the work"
+    assert task.constraints is not None
+    assert "## Constraints" in task.constraints
+    assert "no lint suppressions" in task.constraints
 
 
 async def test_flag_off_attaches_nothing(
@@ -80,20 +83,24 @@ async def test_flag_off_attaches_nothing(
     agent, project = await _seed(db_session)
     task = await TaskService(db_session).create(_req(agent, project, "Do the work"))
     assert task.description == "Do the work"
+    assert task.constraints is None
 
 
 async def test_baseline_not_suppressed_by_agent_constraints_section(
     db_session: AsyncSession, monkeypatch: pytest.MonkeyPatch
 ) -> None:
-    # An agent-authored ## Constraints section must NOT suppress the mandatory
+    # An agent-authored ## Constraints section in the description must NOT
-    # server baseline — both are present.
+    # suppress the mandatory server baseline — the description keeps the
+    # agent's block and `constraints` carries the server baseline (separate
+    # field, so neither suppresses the other).
     monkeypatch.setattr(settings, "conventions_enabled", True)
     agent, project = await _seed(db_session)
     seeded = "Do the work\n\n## Constraints\n- a task-specific note"
     task = await TaskService(db_session).create(_req(agent, project, seeded))
     assert task.description is not None
     assert "a task-specific note" in task.description
-    assert "no lint suppressions" in task.description
+    assert task.constraints is not None
+    assert "no lint suppressions" in task.constraints
 
 
 async def test_baseline_attach_is_idempotent(
@@ -103,8 +110,8 @@ async def test_baseline_attach_is_idempotent(
     agent, project = await _seed(db_session)
     svc = TaskService(db_session)
     task = await svc.create(_req(agent, project, "Do the work"))
-    before = task.description
+    before = task.constraints
     await svc._attach_baseline_constraints(task)
-    assert task.description == before
+    assert task.constraints == before
-    assert task.description is not None
+    assert task.constraints is not None
-    assert task.description.count("no lint suppressions") == 1
+    assert task.constraints.count("no lint suppressions") == 1

tests/unit/api/routes/v1/test_i_will_plan_rich_required.py

@@ -13,7 +13,7 @@ from fastapi import FastAPI
 from fastapi.testclient import TestClient
 from roboco.api.deps import get_choreographer
 from roboco.api.routes.v1.flow_main_pm import router
-from roboco.api.schemas.v1.flow import IWillPlanRequest
+from roboco.api.schemas.v1.flow import IWillPlanRequest, SubTaskCreate
 
 _AGENT_ID = "00000000-0000-0000-0004-000000000001"
 _HEADERS = {"X-Agent-ID": _AGENT_ID, "X-Agent-Role": "main_pm"}
@@ -111,7 +111,9 @@ def test_i_will_plan_schema_accepts_rich_plan() -> None:
         task_id=uuid4(),
         plan="Route to backend",
         approach=_GOOD_APPROACH,
-        sub_tasks=[{"title": "Backend slice", "description": _GOOD_SUBTASK_DESC}],
+        sub_tasks=[
+            SubTaskCreate(title="Backend slice", description=_GOOD_SUBTASK_DESC)
+        ],
         risks=[],
         open_questions=[],
     )

tests/unit/api/test_schemas_v1_flow.py

@@ -11,6 +11,7 @@ from roboco.api.schemas.v1.flow import (
     DelegateRequest,
     IWillPlanRequest,
     IWillWorkOnRequest,
+    SubTaskCreate,
 )
 from roboco.models.base import Complexity
 
@@ -146,3 +147,101 @@ def test_strlist_drops_non_string_junk_instead_of_crashing() -> None:
         technical_considerations=technical_considerations,
     )
     assert req.technical_considerations == ["real note"]
+
+
+# ---------------------------------------------------------------------------
+# Task-content guardrails (2026-07-07): ceilings on plan content + AC caps.
+# ---------------------------------------------------------------------------
+
+
+def test_i_will_plan_request_rejects_overlong_plan() -> None:
+    """plan >2000 chars is rejected at the boundary — the bloat defect."""
+    with pytest.raises(ValidationError) as exc:
+        IWillPlanRequest(
+            task_id=uuid4(),
+            plan="x" * 2001,
+            approach="a" * 150,
+        )
+    assert "plan" in str(exc.value)
+
+
+def test_i_will_plan_request_rejects_overlong_approach() -> None:
+    """approach >800 chars is rejected — no ceiling was the bloat bug."""
+    with pytest.raises(ValidationError) as exc:
+        IWillPlanRequest(
+            task_id=uuid4(),
+            plan="plan",
+            approach="a" * 801,
+        )
+    assert "approach" in str(exc.value)
+
+
+def test_i_will_plan_request_rejects_overlong_subtask_title() -> None:
+    with pytest.raises(ValidationError):
+        IWillPlanRequest(
+            task_id=uuid4(),
+            plan="plan",
+            approach="a" * 150,
+            sub_tasks=[SubTaskCreate(title="t" * 201, description="d" * 30)],
+        )
+
+
+def test_i_will_plan_request_rejects_overlong_subtask_description() -> None:
+    with pytest.raises(ValidationError):
+        IWillPlanRequest(
+            task_id=uuid4(),
+            plan="plan",
+            approach="a" * 150,
+            sub_tasks=[SubTaskCreate(title="ok title", description="d" * 601)],
+        )
+
+
+def test_i_will_plan_request_rejects_thin_subtask_description() -> None:
+    """A sub_task description <20 chars fails the typed SubTaskCreate model."""
+    with pytest.raises(ValidationError):
+        IWillPlanRequest(
+            task_id=uuid4(),
+            plan="plan",
+            approach="a" * 150,
+            sub_tasks=[SubTaskCreate(title="ok title", description="too short")],
+        )
+
+
+def test_delegate_request_rejects_overlong_acceptance_criterion() -> None:
+    """An AC item >200 chars is rejected — a criterion that long is a restated
+    description, not a verifiable outcome."""
+    long_ac = "x" * 201
+    with pytest.raises(ValidationError) as exc:
+        DelegateRequest.model_validate(
+            {
+                "parent_task_id": uuid4(),
+                "title": "t",
+                "description": "add the new endpoint plus tests",
+                "assigned_to": "be-dev-1",
+                "team": "backend",
+                "task_type": "code",
+                "nature": "technical",
+                "estimated_complexity": "medium",
+                "acceptance_criteria": [long_ac],
+            }
+        )
+    assert "acceptance_criteria" in str(exc.value)
+
+
+def test_delegate_request_rejects_too_many_acceptance_criteria() -> None:
+    """An AC list >7 items is rejected — over-decomposition of criteria."""
+    with pytest.raises(ValidationError) as exc:
+        DelegateRequest.model_validate(
+            {
+                "parent_task_id": uuid4(),
+                "title": "t",
+                "description": "add the new endpoint plus tests",
+                "assigned_to": "be-dev-1",
+                "team": "backend",
+                "task_type": "code",
+                "nature": "technical",
+                "estimated_complexity": "medium",
+                "acceptance_criteria": [f"criterion {i}" for i in range(8)],
+            }
+        )
+    assert "acceptance_criteria" in str(exc.value)

tests/unit/gateway/test_i_will_plan_sub_tasks_gate.py

@@ -32,6 +32,8 @@ _GOOD_SUBTASK_DESC = (
     "README H1 leaving the rest untouched, commits with the task-id prefix, "
     "and opens the leaf PR for QA."
 )
+_LONG_SUBTASK_DESC = "x" * 700  # exceeds _PM_SUBTASK_DESC_MAX_LEN (600)
+_OVERLONG_TITLE = "t" * 250  # exceeds _PM_SUBTASK_TITLE_MAX_LEN (200)
 
 
 # ---------------------------------------------------------------------------
@@ -429,3 +431,199 @@ async def test_pm_reentry_in_progress_short_circuits_before_gate() -> None:
         "The gate is firing before _handle_pm_reentry — ordering bug not fixed."
     )
     assert body.get("status") == "in_progress", body
+
+
+# ---------------------------------------------------------------------------
+# Test 7: over-decomposition cap — >7 sub_tasks rejected (2026-07-07 defect)
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_pm_with_too_many_subtasks_gets_incomplete_input() -> None:
+    """A plan with >7 sub_tasks is over-decomposition — rejected with a hint
+    to split into sibling coordination tasks instead of one giant plan."""
+    pm_id = uuid4()
+    task_id = uuid4()
+    task_svc = _pm_task_svc(task_id, role="cell_pm")
+    deps = _make_deps(task=task_svc)
+    c = Choreographer(deps)
+
+    env = await c.i_will_plan(
+        pm_id,
+        task_id,
+        plan="decompose work",
+        rich_plan={
+            "approach": _GOOD_APPROACH,
+            "sub_tasks": [
+                {"title": f"Slice {i}", "description": _GOOD_SUBTASK_DESC}
+                for i in range(8)  # 8 > _PM_SUBTASKS_MAX (7)
+            ],
+        },
+    )
+    body = env.as_dict()
+    assert body["error"] == "incomplete_input", body
+    assert "sub_tasks" in (body.get("missing") or []), body
+    assert "at most 7" in str(body.get("field_hints", {})), body
+
+
+@pytest.mark.asyncio
+async def test_pm_with_seven_subtasks_passes_cap() -> None:
+    """Exactly 7 sub_tasks is the cap — must not be rejected by the cap."""
+    pm_id = uuid4()
+    task_id = uuid4()
+    task_svc = _pm_task_svc(task_id, role="cell_pm")
+    claimed = MagicMock(
+        id=task_id,
+        status="claimed",
+        plan=None,
+        assigned_to=pm_id,
+        task_type="planning",
+    )
+    started = MagicMock(
+        id=task_id,
+        status="in_progress",
+        plan={"text": "x"},
+        assigned_to=pm_id,
+        task_type="planning",
+    )
+    task_svc.claim.return_value = claimed
+    task_svc.set_plan.return_value = claimed
+    task_svc.start.return_value = started
+    deps = _make_deps(task=task_svc)
+    c = Choreographer(deps)
+
+    env = await c.i_will_plan(
+        pm_id,
+        task_id,
+        plan="decompose work",
+        rich_plan={
+            "approach": _GOOD_APPROACH,
+            "sub_tasks": [
+                {"title": f"Slice {i}", "description": _GOOD_SUBTASK_DESC}
+                for i in range(7)  # exactly the cap
+            ],
+        },
+    )
+    body = env.as_dict()
+    assert body.get("error") != "incomplete_input", body
+
+
+# ---------------------------------------------------------------------------
+# Test 8: per-sub_task ceilings — over-long title / description rejected
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_pm_with_overlong_subtask_title_gets_incomplete_input() -> None:
+    """A sub_task title >200 chars is rejected — titles are single-line labels."""
+    pm_id = uuid4()
+    task_id = uuid4()
+    task_svc = _pm_task_svc(task_id, role="cell_pm")
+    deps = _make_deps(task=task_svc)
+    c = Choreographer(deps)
+
+    env = await c.i_will_plan(
+        pm_id,
+        task_id,
+        plan="decompose work",
+        rich_plan={
+            "approach": _GOOD_APPROACH,
+            "sub_tasks": [
+                {"title": _OVERLONG_TITLE, "description": _GOOD_SUBTASK_DESC}
+            ],
+        },
+    )
+    body = env.as_dict()
+    assert body["error"] == "incomplete_input", body
+    assert "sub_tasks" in (body.get("missing") or []), body
+    assert "title is too long" in str(body.get("field_hints", {})), body
+
+
+@pytest.mark.asyncio
+async def test_pm_with_overlong_subtask_description_gets_incomplete_input() -> None:
+    """A sub_task description >600 chars is rejected — over-long sub-task
+    prose is the bloat defect (descriptions dominated by one giant step)."""
+    pm_id = uuid4()
+    task_id = uuid4()
+    task_svc = _pm_task_svc(task_id, role="cell_pm")
+    deps = _make_deps(task=task_svc)
+    c = Choreographer(deps)
+
+    env = await c.i_will_plan(
+        pm_id,
+        task_id,
+        plan="decompose work",
+        rich_plan={
+            "approach": _GOOD_APPROACH,
+            "sub_tasks": [
+                {"title": "Backend slice", "description": _LONG_SUBTASK_DESC}
+            ],
+        },
+    )
+    body = env.as_dict()
+    assert body["error"] == "incomplete_input", body
+    assert "sub_tasks" in (body.get("missing") or []), body
+    assert "description is too long" in str(body.get("field_hints", {})), body
+
+
+# ---------------------------------------------------------------------------
+# Test 9: a code-typed parent WITH sub_tasks is NOT rejected — PMs legitimately
+# plan code-typed parents into dev subtasks (the 2026-05-08 rule). The gate
+# must not ban sub_tasks on code tasks.
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_pm_can_plan_code_typed_parent_with_subtasks() -> None:
+    """A code-typed task planned by a PM with sub_tasks is legitimate
+    decomposition (test_cell_pm_can_plan_code_typed_parent_via_i_will_plan
+    establishes the 2026-05-08 rule). The cap/ceilings gate must NOT reject
+    merely because task_type is 'code'."""
+    pm_id = uuid4()
+    task_id = uuid4()
+    task_svc = AsyncMock()
+    task_svc.get.return_value = MagicMock(
+        id=task_id,
+        status="pending",
+        plan=None,
+        assigned_to=None,
+        task_type="code",
+        parent_task_id=None,
+        sequence=0,
+        team="backend",
+        commits=[],
+        pr_number=None,
+        branch_name=None,
+        quick_context=None,
+    )
+    task_svc.agent_for.return_value = MagicMock(
+        id=pm_id, role="cell_pm", team="backend", slug=None
+    )
+    task_svc.list_in_progress_for_agent.return_value = []
+    task_svc.list_paused_for_agent.return_value = []
+    task_svc.get_subtasks.return_value = []
+    task_svc.session = MagicMock()
+    task_svc.session.begin_nested = MagicMock(
+        return_value=MagicMock(
+            __aenter__=AsyncMock(return_value=None),
+            __aexit__=AsyncMock(return_value=False),
+        )
+    )
+    deps = _make_deps(task=task_svc)
+    c = Choreographer(deps)
+
+    env = await c.i_will_plan(
+        pm_id,
+        task_id,
+        plan="decompose the code parent",
+        rich_plan={
+            "approach": _GOOD_APPROACH,
+            "sub_tasks": [
+                {"title": "API slice", "description": _GOOD_SUBTASK_DESC},
+                {"title": "Test slice", "description": _GOOD_SUBTASK_DESC},
+            ],
+        },
+    )
+    body = env.as_dict()
+    # The gate must not reject merely for task_type='code' with sub_tasks.
+    assert body.get("error") != "incomplete_input", body

uv.lock

@@ -764,18 +764,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/a7/5f/ed01f9a3cdffbd5a008556fc7b2a08ddb1cc6ace7effa7340604b1d16699/docstring_parser-0.18.0-py3-none-any.whl", hash = "sha256:b3fcbed555c47d8479be0796ef7e19c2670d428d72e96da63f3a40122860374b", size = 22484, upload-time = "2026-04-14T04:09:18.638Z" },
 ]
 
-[[package]]
-name = "ecdsa"
-version = "0.19.2"
-source = { registry = "https://pypi.org/simple" }
-dependencies = [
-    { name = "six" },
-]
-sdist = { url = "https://files.pythonhosted.org/packages/25/ca/8de7744cb3bc966c85430ca2d0fcaeea872507c6a4cf6e007f7fe269ed9d/ecdsa-0.19.2.tar.gz", hash = "sha256:62635b0ac1ca2e027f82122b5b81cb706edc38cd91c63dda28e4f3455a2bf930", size = 202432, upload-time = "2026-03-26T09:58:17.675Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/51/79/119091c98e2bf49e24ed9f3ae69f816d715d2904aefa6a2baa039a2ba0b0/ecdsa-0.19.2-py2.py3-none-any.whl", hash = "sha256:840f5dc5e375c68f36c1a7a5b9caad28f95daa65185c9253c0c08dd952bb7399", size = 150818, upload-time = "2026-03-26T09:58:15.808Z" },
-]
-
 [[package]]
 name = "email-validator"
 version = "2.3.0"
@@ -2113,15 +2101,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/9b/bf/7595e817906a29453ba4d99394e781b6fabe55d21f3c15d240f85dd06bb1/py_serializable-2.1.0-py3-none-any.whl", hash = "sha256:b56d5d686b5a03ba4f4db5e769dc32336e142fc3bd4d68a8c25579ebb0a67304", size = 23045, upload-time = "2025-07-21T09:56:46.848Z" },
 ]
 
-[[package]]
-name = "pyasn1"
-version = "0.6.3"
-source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/5c/5f/6583902b6f79b399c9c40674ac384fd9cd77805f9e6205075f828ef11fb2/pyasn1-0.6.3.tar.gz", hash = "sha256:697a8ecd6d98891189184ca1fa05d1bb00e2f84b5977c481452050549c8a72cf", size = 148685, upload-time = "2026-03-17T01:06:53.382Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/5d/a0/7d793dce3fa811fe047d6ae2431c672364b462850c6235ae306c0efd025f/pyasn1-0.6.3-py3-none-any.whl", hash = "sha256:a80184d120f0864a52a073acc6fc642847d0be408e7c7252f31390c0f4eadcde", size = 83997, upload-time = "2026-03-17T01:06:52.036Z" },
-]
-
 [[package]]
 name = "pycparser"
 version = "3.0"
@@ -2342,25 +2321,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/0b/d7/1959b9648791274998a9c3526f6d0ec8fd2233e4d4acce81bbae76b44b2a/python_dotenv-1.2.2-py3-none-any.whl", hash = "sha256:1d8214789a24de455a8b8bd8ae6fe3c6b69a5e3d64aa8a8e5d68e694bbcb285a", size = 22101, upload-time = "2026-03-01T16:00:25.09Z" },
 ]
 
-[[package]]
-name = "python-jose"
-version = "3.5.0"
-source = { registry = "https://pypi.org/simple" }
-dependencies = [
-    { name = "ecdsa" },
-    { name = "pyasn1" },
-    { name = "rsa" },
-]
-sdist = { url = "https://files.pythonhosted.org/packages/c6/77/3a1c9039db7124eb039772b935f2244fbb73fc8ee65b9acf2375da1c07bf/python_jose-3.5.0.tar.gz", hash = "sha256:fb4eaa44dbeb1c26dcc69e4bd7ec54a1cb8dd64d3b4d81ef08d90ff453f2b01b", size = 92726, upload-time = "2025-05-28T17:31:54.288Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/d9/c3/0bd11992072e6a1c513b16500a5d07f91a24017c5909b02c72c62d7ad024/python_jose-3.5.0-py2.py3-none-any.whl", hash = "sha256:abd1202f23d34dfad2c3d28cb8617b90acf34132c7afd60abd0b0b7d3cb55771", size = 34624, upload-time = "2025-05-28T17:31:52.802Z" },
-]
-
-[package.optional-dependencies]
-cryptography = [
-    { name = "cryptography" },
-]
-
 [[package]]
 name = "python-multipart"
 version = "0.0.32"
@@ -2600,7 +2560,6 @@ dependencies = [
     { name = "pydantic" },
     { name = "pydantic-settings" },
     { name = "pyjwt" },
-    { name = "python-jose", extra = ["cryptography"] },
     { name = "python-multipart" },
     { name = "python-toon" },
     { name = "pyyaml" },
@@ -2637,7 +2596,6 @@ dev = [
     { name = "rich" },
     { name = "ruff" },
     { name = "types-passlib" },
-    { name = "types-python-jose" },
     { name = "types-pyyaml" },
     { name = "vulture" },
     { name = "xenon" },
@@ -2685,7 +2643,6 @@ requires-dist = [
     { name = "pytest-asyncio", marker = "extra == 'dev'" },
     { name = "pytest-cov", marker = "extra == 'dev'" },
     { name = "pytest-xdist", marker = "extra == 'dev'" },
-    { name = "python-jose", extras = ["cryptography"] },
     { name = "python-multipart" },
     { name = "python-toon" },
     { name = "pyyaml" },
@@ -2703,7 +2660,6 @@ requires-dist = [
     { name = "tree-sitter-python" },
     { name = "tree-sitter-typescript" },
     { name = "types-passlib", marker = "extra == 'dev'" },
-    { name = "types-python-jose", marker = "extra == 'dev'" },
     { name = "types-pyyaml", marker = "extra == 'dev'" },
     { name = "uvicorn", extras = ["standard"] },
     { name = "vulture", marker = "extra == 'dev'" },
@@ -2771,18 +2727,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/a6/b2/9574d4d44f7760c2aa32d92a0a4f41698e33f5b204a0bf5c9758f52c79d5/rpds_py-2026.6.3-cp314-cp314t-win_amd64.whl", hash = "sha256:c7b9a2f8f4d8e90af72571d3d495deebdd7e3c75451f5b41719aee166e940fc2", size = 220600, upload-time = "2026-06-30T07:16:43.091Z" },
 ]
 
-[[package]]
-name = "rsa"
-version = "4.9.1"
-source = { registry = "https://pypi.org/simple" }
-dependencies = [
-    { name = "pyasn1" },
-]
-sdist = { url = "https://files.pythonhosted.org/packages/da/8a/22b7beea3ee0d44b1916c0c1cb0ee3af23b700b6da9f04991899d0c555d4/rsa-4.9.1.tar.gz", hash = "sha256:e7bdbfdb5497da4c07dfd35530e1a902659db6ff241e39d9953cad06ebd0ae75", size = 29034, upload-time = "2025-04-16T09:51:18.218Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/64/8d/0133e4eb4beed9e425d9a98ed6e081a55d195481b7632472be1af08d2f6b/rsa-4.9.1-py3-none-any.whl", hash = "sha256:68635866661c6836b8d39430f97a996acbd61bfa49406748ea243539fe239762", size = 34696, upload-time = "2025-04-16T09:51:17.142Z" },
-]
-
 [[package]]
 name = "ruff"
 version = "0.15.20"
@@ -3110,27 +3054,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/14/6a/e9fc6a5b8f9a380a4a56b9f1e4dba5c6899561868017b17f6de382808b6f/types_passlib-1.7.7.20260211-py3-none-any.whl", hash = "sha256:c0f1ad440c513a6c07f333b28249530686056fd54a7b3ac6128ae31fd46305d3", size = 40457, upload-time = "2026-02-10T15:11:58.647Z" },
 ]
 
-[[package]]
-name = "types-pyasn1"
-version = "0.6.0.20260408"
-source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/58/c0/02f897fc8543f64fa6b1ca6a30d388e37c4ec2f761f469a2d9a29b89cdef/types_pyasn1-0.6.0.20260408.tar.gz", hash = "sha256:32dc90927adbe504fd2eee83ae30cf5ef934e5db0d1d94886071fed47eb50c8c", size = 17312, upload-time = "2026-04-08T04:27:16.874Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/e1/a5/473e06d5aaec3730aab5a9d40c2044e673c927412c24bd7f3fa0df7e95d3/types_pyasn1-0.6.0.20260408-py3-none-any.whl", hash = "sha256:ee7fbd98bce61193c5d4f8f7812fa53cddc5b8cc5ceb9fcda6eea539947c6d6b", size = 24044, upload-time = "2026-04-08T04:27:16.002Z" },
-]
-
-[[package]]
-name = "types-python-jose"
-version = "3.5.0.20260408"
-source = { registry = "https://pypi.org/simple" }
-dependencies = [
-    { name = "types-pyasn1" },
-]
-sdist = { url = "https://files.pythonhosted.org/packages/a1/b1/87cdb410d22913df39d7aad8864f94a4f5ad8d507ee07556888fdbe55e19/types_python_jose-3.5.0.20260408.tar.gz", hash = "sha256:3f8dccdc327bfffea7a81084ea1cea722fa499f13c1d04f7978b491dd36e0cf1", size = 11989, upload-time = "2026-04-08T04:34:10.577Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/d3/83/df2b34e64f0a674935d718471cf10fb392a7e5bdb0e9e7c739885b62d274/types_python_jose-3.5.0.20260408-py3-none-any.whl", hash = "sha256:968d8a8eac1ff9da249d6335a2bb9f82288d59ba23afe91fcc2662eb9f485e2a", size = 14694, upload-time = "2026-04-08T04:34:09.747Z" },
-]
-
 [[package]]
 name = "types-pyyaml"
 version = "6.0.12.20260518"