@event4u/agent-config 1.12.0 → 1.14.0

package/.agent-src/templates/scripts/work_engine/directives/backend/__init__.py

@@ -0,0 +1,98 @@
+"""Backend directive set — step handlers for the backend-coding flow.
+
+Each module exposes a single ``run`` callable that matches the
+``Step`` protocol defined in ``...delivery_state``. The dispatcher
+wires them into the ``STEP_ORDER`` mapping at call time; nothing in
+this package imports handlers eagerly, so a partial wiring during
+development is caught by the dispatcher's missing-step check rather
+than by an import-time failure.
+
+This is the first concrete entry in the
+:mod:`work_engine.directives` package. R1 Phase 4 Step 3 moved the
+handlers here from the original ``work_engine.steps`` location so
+the upcoming generalized dispatcher (Step 2) can select between
+multiple directive sets uniformly via :func:`get_steps`. The
+external behavior — flow order, ambiguity surfaces, halt-points —
+is unchanged.
+
+The deterministic gates (``refine``, ``memory``, ``analyze``)
+validate upstream state; the delegation gates (``plan``,
+``implement``, ``test``, ``verify``) halt with
+``@agent-directive:`` markers so the orchestrator can invoke the
+matching skill and resume. ``report`` renders the delivery Markdown
+once everything else has succeeded. See
+``agents/roadmaps/road-to-implement-ticket.md`` for the shipping
+order and ``agents/contexts/implement-ticket-flow.md`` for the
+slice contracts each handler writes to.
+"""
+from __future__ import annotations
+
+from collections.abc import Mapping
+
+from ...delivery_state import Step
+from . import analyze, implement, memory, plan, refine, report, test, verify
+
+DIRECTIVE_SET_NAME = "backend"
+"""External name carried in ``state.directive_set`` for this set."""
+
+SUPPORTED_KINDS: tuple[str, ...] = ("ticket", "prompt")
+"""Input kinds this directive set knows how to handle.
+
+Read by :func:`work_engine.dispatcher.assert_kind_supported` before the
+loop starts. The schema's :data:`work_engine.state.KNOWN_INPUT_KINDS` is
+the *envelope* whitelist (what is accepted on disk); ``SUPPORTED_KINDS``
+is the *capability* whitelist (what this set can actually drive end to
+end).
+
+R1 only carried ``ticket``. R2 Phase 3 Step 3 added ``prompt`` once the
+``refine`` step learned to detect prompt envelopes (presence of ``raw``
+key in ``state.ticket``), delegate to the ``refine-prompt`` skill on
+the first pass, and route the resulting confidence band into
+``SUCCESS`` / ``PARTIAL`` / ``BLOCKED`` per
+``agents/roadmaps/archive/road-to-prompt-driven-execution.md`` Phase 3."""
+
+_STEPS = (refine, memory, analyze, plan, implement, test, verify, report)
+
+
+def all_ambiguities() -> dict[str, tuple[dict[str, str], ...]]:
+    """Return `{step_name: AMBIGUITIES}` for every step in flow order.
+
+    Used by documentation generators and the ``test_ambiguity_coverage``
+    suite to prove every step explicitly declares what can surface a
+    ``BLOCKED`` outcome. Steps that always succeed (``memory``,
+    ``report``) return an empty tuple — declared intent, not an
+    omission.
+    """
+    return {step.__name__.rsplit(".", 1)[-1]: step.AMBIGUITIES for step in _STEPS}
+
+
+def get_steps() -> Mapping[str, Step]:
+    """Return the ``{step_name: handler}`` mapping the dispatcher walks.
+
+    Each value is the module-level ``run`` callable matching the
+    :data:`work_engine.delivery_state.Step` protocol —
+    ``Callable[[DeliveryState], StepResult]`` — exactly what
+    :func:`work_engine.dispatcher.dispatch` calls. Order of insertion
+    matches the canonical backend flow (refine → memory → analyze →
+    plan → implement → test → verify → report); the dispatcher's own
+    ``STEP_ORDER`` is the single source of truth for *which* steps
+    exist, but the directive set is the single source of truth for
+    *how* each one runs.
+    """
+    return {step.__name__.rsplit(".", 1)[-1]: step.run for step in _STEPS}
+
+
+__all__ = [
+    "DIRECTIVE_SET_NAME",
+    "SUPPORTED_KINDS",
+    "all_ambiguities",
+    "analyze",
+    "get_steps",
+    "implement",
+    "memory",
+    "plan",
+    "refine",
+    "report",
+    "test",
+    "verify",
+]

package/.agent-src/templates/scripts/{implement_ticket/steps → work_engine/directives/backend}/analyze.py

@@ -20,7 +20,7 @@ rule. Otherwise it returns ``SUCCESS`` without mutating state.
 """
 from __future__ import annotations
 
-from ..delivery_state import DeliveryState, Outcome, StepResult
+from ...delivery_state import DeliveryState, Outcome, StepResult
 
 AMBIGUITIES: tuple[dict[str, str], ...] = (
     {

package/.agent-src/templates/scripts/{implement_ticket/steps → work_engine/directives/backend}/implement.py

@@ -25,13 +25,13 @@ from __future__ import annotations
 
 from typing import Any
 
-from ..delivery_state import (
+from ...delivery_state import (
     DeliveryState,
     Outcome,
     StepResult,
     agent_directive,
 )
-from ..persona_policy import resolve_policy
+from ...persona_policy import resolve_policy
 
 AMBIGUITIES: tuple[dict[str, str], ...] = (
     {

package/.agent-src/templates/scripts/{implement_ticket/steps → work_engine/directives/backend}/memory.py

@@ -23,7 +23,7 @@ from __future__ import annotations
 import re
 from typing import Any, Iterable
 
-from ..delivery_state import DeliveryState, Outcome, StepResult
+from ...delivery_state import DeliveryState, Outcome, StepResult
 
 MEMORY_TYPES: tuple[str, ...] = (
     "domain-invariants",

package/.agent-src/templates/scripts/{implement_ticket/steps → work_engine/directives/backend}/plan.py

@@ -28,7 +28,7 @@ from __future__ import annotations
 
 from typing import Any
 
-from ..delivery_state import (
+from ...delivery_state import (
     DeliveryState,
     Outcome,
     StepResult,

package/.agent-src/templates/scripts/work_engine/directives/backend/refine.py

@@ -0,0 +1,396 @@
+"""``refine`` step — deterministic gate in front of the refinement skills.
+
+The step never calls an LLM. It inspects ``state.ticket`` (which carries
+``input.data`` after the CLI projection) and routes on shape:
+
+- **Ticket envelope** (``id``, ``title``, ``acceptance_criteria``) — the
+  R1 path. Validates the minimum viable shape and either returns
+  ``SUCCESS`` or ``BLOCKED`` with numbered options pointing at
+  ``/refine-ticket``.
+- **Prompt envelope** (``raw`` key present, ``reconstructed_ac`` /
+  ``assumptions`` slots) — the R2 path. On the first pass the gate
+  delegates to the ``refine-prompt`` skill via an ``@agent-directive:``
+  halt; on the rebound it scores the reconstructed envelope and routes
+  the resulting confidence band:
+
+  - ``high``   → ``SUCCESS`` (silent proceed, breakdown logged for the report)
+  - ``medium`` → ``PARTIAL`` with an assumptions-report halt
+  - ``low``    → ``BLOCKED`` with one clarifying question targeted at the
+    weakest dimension
+
+The checks live here (rather than inside the refinement skills) because
+the dispatcher is synchronous Python: it cannot "delegate" to an agent
+skill mid-loop. Making the gate deterministic keeps the contract "block
+on ambiguity, never guess" enforceable from code, and ensures the band
+the dispatcher routes on is always engine-computed — the skill produces
+AC + assumptions, the engine decides.
+"""
+from __future__ import annotations
+
+from typing import Any
+
+from ...delivery_state import (
+    DeliveryState,
+    Outcome,
+    StepResult,
+    agent_directive,
+)
+from ...scoring import confidence as _confidence
+
+_MIN_TITLE_LEN = 3
+_MIN_AC_LEN = 10
+
+AMBIGUITIES: tuple[dict[str, str], ...] = (
+    {
+        "code": "missing_id",
+        "trigger": "ticket has no `id` field (or only whitespace)",
+        "resolution": "run `/refine-ticket` or paste the ticket id in chat",
+    },
+    {
+        "code": "trivial_title",
+        "trigger": f"title missing or shorter than {_MIN_TITLE_LEN} chars",
+        "resolution": "run `/refine-ticket` to rewrite the title",
+    },
+    {
+        "code": "missing_or_vague_ac",
+        "trigger": (
+            f"acceptance_criteria empty, non-list, or any item under "
+            f"{_MIN_AC_LEN} chars"
+        ),
+        "resolution": "run `/refine-ticket` to add concrete acceptance criteria",
+    },
+    {
+        "code": "prompt_unrefined",
+        "trigger": (
+            "prompt envelope present but `reconstructed_ac` is empty — "
+            "the deterministic gate has nothing to score yet"
+        ),
+        "resolution": "agent directive `refine-prompt` → run the skill, "
+        "write AC + assumptions back into `state.ticket`",
+    },
+    {
+        "code": "prompt_medium_confidence",
+        "trigger": (
+            "scored band is `medium` and the user has not confirmed the "
+            "assumptions report yet"
+        ),
+        "resolution": "user confirms the reconstructed AC + assumptions, "
+        "or refines them; agent flips `confidence_confirmed=True` to "
+        "release the gate",
+    },
+    {
+        "code": "prompt_low_confidence",
+        "trigger": (
+            "scored band is `low` — too little signal to plan against, "
+            "even after reconstruction"
+        ),
+        "resolution": "user answers one clarifying question; the agent "
+        "re-runs `refine-prompt` against the refreshed prompt",
+    },
+    {
+        "code": "prompt_ui_intent",
+        "trigger": (
+            "scorer flagged `ui_intent=True` — the prompt reads as UI "
+            "work and the backend track cannot ship it cleanly"
+        ),
+        "resolution": "user re-frames the prompt as backend-only, parks "
+        "it for Roadmap 3 (`road-to-product-ui-track.md`), or aborts",
+    },
+)
+"""Declared ambiguity surfaces. Every BLOCKED / PARTIAL return maps to one code."""
+
+
+def run(state: DeliveryState) -> StepResult:
+    """Route on envelope shape: ticket path or prompt path."""
+    data = state.ticket or {}
+    if _is_prompt_envelope(data):
+        return _run_prompt(state, data)
+
+    deficiencies = _diagnose(data)
+    if not deficiencies:
+        return StepResult(outcome=Outcome.SUCCESS)
+
+    ticket_id = data.get("id") or "(no id)"
+    questions = _format_questions(ticket_id, deficiencies)
+    return StepResult(
+        outcome=Outcome.BLOCKED,
+        questions=questions,
+        message=(
+            f"Ticket {ticket_id} is not refined enough to plan against: "
+            + "; ".join(deficiencies)
+        ),
+    )
+
+
+def _is_prompt_envelope(data: dict[str, Any]) -> bool:
+    """True when ``state.ticket`` carries a prompt envelope.
+
+    The presence of a string-valued ``raw`` key is unambiguous: ticket
+    payloads never carry ``raw``, and prompt envelopes always do (the
+    resolver writes it before any handler sees the state).
+    """
+    if not isinstance(data, dict):
+        return False
+    raw = data.get("raw")
+    return isinstance(raw, str) and bool(raw.strip())
+
+
+def _diagnose(ticket: dict[str, Any]) -> list[str]:
+    """Return a human-readable list of what's missing from the ticket.
+
+    Order matches what a reader needs first (identity → summary →
+    acceptance criteria) so the surfaced questions read naturally.
+    """
+    issues: list[str] = []
+
+    ticket_id = ticket.get("id")
+    if not isinstance(ticket_id, str) or not ticket_id.strip():
+        issues.append("missing ticket id")
+
+    title = ticket.get("title")
+    if not isinstance(title, str) or len(title.strip()) < _MIN_TITLE_LEN:
+        issues.append("missing or trivial title")
+
+    ac = ticket.get("acceptance_criteria")
+    if not isinstance(ac, list) or not ac:
+        issues.append("no acceptance criteria")
+    else:
+        weak_indices = [
+            idx + 1
+            for idx, item in enumerate(ac)
+            if not _is_concrete_ac(item)
+        ]
+        if weak_indices:
+            issues.append(
+                "vague acceptance criteria at position(s) "
+                + ", ".join(str(i) for i in weak_indices),
+            )
+
+    return issues
+
+
+def _is_concrete_ac(item: Any) -> bool:
+    """An AC is concrete when it is a non-empty string above the length floor.
+
+    The floor is deliberately loose: refine is a gate, not a style
+    judge. The heavy lifting (measurability, testability, tone) is
+    owned by the ``refine-ticket`` skill on the rebound.
+    """
+    if not isinstance(item, str):
+        return False
+    return len(item.strip()) >= _MIN_AC_LEN
+
+
+def _format_questions(ticket_id: str, deficiencies: list[str]) -> list[str]:
+    """Render the numbered options shown to the user when BLOCKED.
+
+    Three options, ordered by likely next action: run the existing
+    refinement skill, paste the missing data in chat, or abandon the
+    ticket entirely. ``user-interaction`` requires numbered, prose-
+    free options; the deficiency list is rendered as a headnote.
+    """
+    headnote = (
+        "> Ticket "
+        + ticket_id
+        + " is missing: "
+        + "; ".join(deficiencies)
+        + "."
+    )
+    return [
+        headnote,
+        f"> 1. Run `/refine-ticket {ticket_id}` and re-invoke `/implement-ticket`",
+        "> 2. Provide the missing details in chat — I'll merge them into the ticket",
+        "> 3. Abandon this ticket — too vague to implement",
+    ]
+
+
+def _run_prompt(state: DeliveryState, data: dict[str, Any]) -> StepResult:
+    """Score the prompt envelope and route on the resulting band.
+
+    First pass (no AC reconstructed yet) → delegate to ``refine-prompt``.
+    Second pass → score and branch:
+
+    - ``high``   → ``SUCCESS``; the breakdown is recorded on
+      ``state.ticket['confidence']`` so the report renderer can include
+      it without re-scoring.
+    - ``medium`` → ``PARTIAL`` with an assumptions-report halt unless
+      the agent has flipped ``confidence_confirmed=True`` after the
+      user signed off. ``low`` band can never be released this way.
+    - ``low``    → ``BLOCKED`` with one clarifying question targeted at
+      the weakest dimension (lowest score wins; ties prefer the order
+      declared in :data:`work_engine.scoring.confidence.DIMENSION_NAMES`).
+    """
+    raw = data.get("raw") or ""
+    ac = data.get("reconstructed_ac")
+    if not isinstance(ac, list):
+        ac = []
+    assumptions = data.get("assumptions")
+    if not isinstance(assumptions, list):
+        assumptions = []
+
+    if not ac:
+        return _delegate_to_refine_prompt(raw)
+
+    result = _confidence.score(raw=raw, ac=ac, assumptions=assumptions)
+    data["confidence"] = {
+        "band": result.band,
+        "score": result.score,
+        "dimensions": dict(result.dimensions),
+        "reasons": list(result.reasons),
+        "ui_intent": result.ui_intent,
+    }
+    # Mirror reconstructed AC into the legacy slot every downstream gate
+    # (analyze, plan, implement) reads. Prompt envelopes carry AC under
+    # ``reconstructed_ac``; without this projection ``analyze`` blocks with
+    # "ticket lost its acceptance criteria" the moment ``refine`` succeeds.
+    data["acceptance_criteria"] = list(ac)
+
+    if result.ui_intent:
+        return _halt_ui_intent(raw, result)
+
+    if result.band == "high":
+        return StepResult(outcome=Outcome.SUCCESS)
+
+    if result.band == "medium":
+        if data.get("confidence_confirmed") is True:
+            return StepResult(outcome=Outcome.SUCCESS)
+        return _halt_medium(raw, ac, assumptions, result)
+
+    return _halt_low(raw, result)
+
+
+def _delegate_to_refine_prompt(raw: str) -> StepResult:
+    """Halt with an agent directive so the orchestrator runs ``refine-prompt``."""
+    preview = _preview(raw)
+    return StepResult(
+        outcome=Outcome.BLOCKED,
+        questions=[
+            agent_directive("refine-prompt"),
+            f"> Prompt received: {preview}",
+            "> No reconstructed acceptance criteria yet — running "
+            "`refine-prompt` and resuming.",
+            "> 1. Continue — let the skill reconstruct AC + assumptions",
+            "> 2. Abort — the prompt is not what I meant",
+        ],
+        message="Prompt envelope present but unrefined; delegating to refine-prompt.",
+    )
+
+
+def _halt_medium(
+    raw: str,
+    ac: list[Any],
+    assumptions: list[Any],
+    result: "_confidence.ConfidenceScore",
+) -> StepResult:
+    """PARTIAL halt — assumptions report, one user round-trip."""
+    preview = _preview(raw)
+    ac_lines = [f">    {idx}. {item}" for idx, item in enumerate(ac, start=1)]
+    asm_lines = [f">    - {item}" for item in assumptions] or [
+        ">    - (none recorded)",
+    ]
+    questions = [
+        f"> Prompt: {preview}",
+        f"> Confidence: **medium** (score {result.score:.2f}). "
+        "Assumptions worth confirming before I plan.",
+        "> Reconstructed AC:",
+        *ac_lines,
+        "> Assumptions:",
+        *asm_lines,
+        "> 1. Continue as-is — the AC + assumptions are good enough",
+        "> 2. Refine — I'll send a corrected prompt and re-run "
+        "`refine-prompt`",
+        "> 3. Abort — pause this `/work` cycle",
+    ]
+    return StepResult(
+        outcome=Outcome.PARTIAL,
+        questions=questions,
+        message=(
+            f"Prompt scored medium ({result.score:.2f}); "
+            "halting for assumptions confirmation."
+        ),
+    )
+
+
+def _halt_low(raw: str, result: "_confidence.ConfidenceScore") -> StepResult:
+    """BLOCKED halt — one targeted question on the weakest dimension."""
+    preview = _preview(raw)
+    weakest_idx, weakest_name = _weakest_dimension(result.dimensions)
+    reason = result.reasons[weakest_idx] if weakest_idx < len(result.reasons) else ""
+    prompts = {
+        "goal_clarity": "What is the single observable outcome you want?",
+        "scope_boundary": "Which file, class, or module should I touch?",
+        "ac_evidence": "What concrete behaviour proves it works?",
+        "stack_data": "Which table, column, or migration target is involved?",
+        "reversibility": "Is this change destructive — should I work behind a flag?",
+    }
+    question = prompts.get(weakest_name, "Can you tighten the prompt?")
+    return StepResult(
+        outcome=Outcome.BLOCKED,
+        questions=[
+            f"> Prompt: {preview}",
+            f"> Confidence: **low** (score {result.score:.2f}). "
+            f"Weakest dimension: `{weakest_name}` — {reason}",
+            f"> {question}",
+            "> 1. I'll answer — paste the answer in chat and re-invoke `/work`",
+            "> 2. Abort — drop this prompt",
+        ],
+        message=(
+            f"Prompt scored low ({result.score:.2f}); blocking on "
+            f"`{weakest_name}` clarification."
+        ),
+    )
+
+
+def _halt_ui_intent(
+    raw: str, result: "_confidence.ConfidenceScore",
+) -> StepResult:
+    """BLOCKED halt — UI-shaped prompts await the R3 dispatch track.
+
+    The backend `directives/backend/` set has no UI capability; routing a
+    UI prompt through it would either ship a backend stub or guess at a
+    component. Both are worse than a clean refusal with a pointer to the
+    deferred R3 track. The halt is band-independent — even a high-band
+    UI prompt blocks here, because confidence on the *reconstruction* says
+    nothing about whether the *dispatcher* can deliver it.
+    """
+    preview = _preview(raw)
+    return StepResult(
+        outcome=Outcome.BLOCKED,
+        questions=[
+            f"> Prompt: {preview}",
+            "> This prompt reads as **UI work** — the backend dispatch "
+            "track can't ship it cleanly.",
+            "> UI dispatch is deferred to Roadmap 3 "
+            "(`road-to-product-ui-track.md`); until it lands, `/work` "
+            "only handles backend-shaped prompts.",
+            "> 1. Re-frame as a backend-only prompt — I'll re-score and proceed",
+            "> 2. Park this prompt — wait for R3 and re-invoke `/work` then",
+            "> 3. Abort — drop this prompt",
+        ],
+        message=(
+            f"Prompt flagged as UI-intent (band={result.band}, "
+            f"score={result.score:.2f}); blocked pending R3 UI track."
+        ),
+    )
+
+
+def _weakest_dimension(dimensions: dict[str, int]) -> tuple[int, str]:
+    """Return ``(index, name)`` of the lowest-scoring dimension.
+
+    Ties are broken by :data:`_confidence.DIMENSION_NAMES` order so the
+    same input always produces the same question (replay determinism).
+    """
+    ordered = list(_confidence.DIMENSION_NAMES)
+    weakest_name = min(ordered, key=lambda n: (dimensions.get(n, 0), ordered.index(n)))
+    return ordered.index(weakest_name), weakest_name
+
+
+def _preview(raw: str, max_chars: int = 80) -> str:
+    """Trim a raw prompt for inline display in halts."""
+    text = " ".join((raw or "").split())
+    if len(text) <= max_chars:
+        return text
+    return text[: max_chars - 1].rstrip() + "\u2026"
+
+
+__all__ = ["AMBIGUITIES", "run"]

package/.agent-src/templates/scripts/{implement_ticket/steps → work_engine/directives/backend}/report.py

@@ -17,8 +17,8 @@ from __future__ import annotations
 
 from typing import Any, Iterable
 
-from ..delivery_state import DeliveryState, Outcome, StepResult
-from ..persona_policy import resolve_policy
+from ...delivery_state import DeliveryState, Outcome, StepResult
+from ...persona_policy import resolve_policy
 
 AMBIGUITIES: tuple[dict[str, str], ...] = ()
 """Report rendering is pure and always succeeds — no blocked paths."""
@@ -38,12 +38,13 @@ def _render(state: DeliveryState) -> str:
         _changes_section(state),
         _tests_section(state),
         _verify_section(state),
+        _visual_preview_section(state),
         _memory_section(state),
         _followups_section(state),
         _next_commands_section(state),
     ]
-    # Drop sections that opted out (memory-that-mattered returns "" when no
-    # hit influenced an outcome — per the report schema drop-rule).
+    # Drop sections that opted out (memory-that-mattered and visual-preview
+    # return "" when their slice is absent — per the report schema drop-rule).
     return "\n\n".join(section for section in sections if section)
 
 
@@ -117,6 +118,37 @@ def _verify_section(state: DeliveryState) -> str:
     return "## Verify\n\n" + _format_kv_block(state.verify, "_(no verify verdict)_")
 
 
+def _visual_preview_section(state: DeliveryState) -> str:
+    """R4 Phase 3: render captured preview artifacts when the skill rendered.
+
+    Reads ``state.ui_review.preview`` (engine never renders — the
+    stack-specific review skill writes the envelope). Emits a section
+    only when ``render_ok`` is ``True`` AND at least one artifact path
+    is present. Failed renders, skipped previews, and pre-R4 envelopes
+    drop the whole section (heading included).
+    """
+    ui_review = getattr(state, "ui_review", None)
+    if not isinstance(ui_review, dict):
+        return ""
+    preview = ui_review.get("preview")
+    if not isinstance(preview, dict):
+        return ""
+    if preview.get("render_ok") is not True:
+        return ""
+    if preview.get("skipped"):
+        return ""
+    screenshot = preview.get("screenshot_path")
+    dom_dump = preview.get("dom_dump_path")
+    lines: list[str] = []
+    if isinstance(screenshot, str) and screenshot:
+        lines.append(f"- Screenshot: `{screenshot}`")
+    if isinstance(dom_dump, str) and dom_dump:
+        lines.append(f"- DOM dump: `{dom_dump}`")
+    if not lines:
+        return ""
+    return "\n".join(["## Visual preview", "", *lines])
+
+
 def _memory_section(state: DeliveryState) -> str:
     """Render **only** hits that changed an outcome (per report schema)."""
     influential = [

package/.agent-src/templates/scripts/{implement_ticket/steps → work_engine/directives/backend}/test.py

@@ -22,13 +22,13 @@ from __future__ import annotations
 
 from typing import Any
 
-from ..delivery_state import (
+from ...delivery_state import (
     DeliveryState,
     Outcome,
     StepResult,
     agent_directive,
 )
-from ..persona_policy import resolve_policy
+from ...persona_policy import resolve_policy
 
 _ALLOWED_VERDICTS = ("success", "failed", "mixed")
 

package/.agent-src/templates/scripts/{implement_ticket/steps → work_engine/directives/backend}/verify.py

@@ -22,13 +22,13 @@ from __future__ import annotations
 
 from typing import Any
 
-from ..delivery_state import (
+from ...delivery_state import (
     DeliveryState,
     Outcome,
     StepResult,
     agent_directive,
 )
-from ..persona_policy import resolve_policy
+from ...persona_policy import resolve_policy
 
 _ALLOWED_VERDICTS = ("success", "blocked", "partial")