Fixes from the first live session of the new loop

Found live on 2026-10-07 (bring-up on nisse-testing-operator/nisse-bringup,
Opus as driver):

- The driver's prose nudge went into the system prompt, so the retry ended
  on the model's own reply. Opus 5.5 refuses that ("does not support
  assistant message prefill. The conversation must end with a user
  message"). The nudge is now a harness fact in the conversation. The
  delegation harness's prose path did the same and is fixed likewise.
- A helper's first request ended with its own delegation_started event,
  rendered as an assistant message: every delegation would have failed on
  Opus. The helper now sees it as the harness handing over its task.
- model_contract checks that every request ends on the user's side, so
  scripted tests catch this from now on (each fix is mutation-checked).
- `status` (and `details`) in an item's thread were taken as replies: they
  woke the driver and cancelled a bug's countdown. They now answer at once
  and change nothing.
- The driver had no way to answer a remark without changing the item, so it
  answered in prose. Every run woken by a message (or a review comment) now
  offers `reply`, which answers where asked and returns to the very wait
  that was interrupted.

Also: scripts/bringup-repo.sh makes the bring-up's test repository
(make_repo/delete_repo, split out of fresh_repo), and made repositories get
the agent's labels (bug, discuss, design).

Tests: test_loop_live_fixes.py (one per finding); the walk sends `status`
into threads (status_changes_nothing) and its scripted driver sometimes
answers remarks with reply.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPQ14L3LikyefCMFBqEbKE

nisse-dev/src/nisse_dev/delegation.py

@@ -33,6 +33,7 @@ from nisse_dev.canonical import (
     AssistantText,
     DelegationCompleted,
     DelegationStarted,
+    Fact,
     PlanSubmitted,
     StoredEvent,
     ToolCall,
@@ -414,6 +415,15 @@ class Delegation:
                 # and its tree state is unknown. Treat it as a failure rather
                 # than inventing a handback on its behalf.
                 if any(isinstance(e, AssistantText) for e in produced):
+                    # Said something but did nothing: told so, as the
+                    # harness, so the next request ends on the user's side
+                    # (models refuse assistant prefill).
+                    transcript.as_actor("orchestrator").append(
+                        Fact(
+                            text="Your last reply was prose. Use your tools to "
+                            "do the work, then call hand_back."
+                        )
+                    )
                     continue
                 return DelegationResult(
                     "failed",

nisse-dev/src/nisse_dev/loop/agent.py

@@ -116,6 +116,10 @@ asking is better than guessing.
 Before each group of tool calls, say in a sentence what you're doing and why:
 the operator sees it as your progress.
 
+When the operator says something that needs an answer but no change (a
+remark, a status question), answer with reply: the item goes back to what it
+was waiting for. Every reply of yours is a tool call, never bare prose.
+
 The project's documents are in its repository: read ARCHITECTURE.md and any
 build plans under docs/plans/ before you restate.
 
@@ -587,6 +591,8 @@ class LoopAgent:
                 loop, position = replace(loop, pr=0, pr_url="", branch=""), READY
             elif new and item.position == WAITING:
                 position = READY
+                # A comment can be answered with a reply, back to this wait.
+                loop = replace(loop, back_to=loop.wait, back_note=loop.note)
             if position != item.position or loop != loop_of(item):
                 if position == READY:
                     loop = replace(loop, wait="", note="")
@@ -824,6 +830,19 @@ class LoopAgent:
             elif not self._store.active_item():
                 self._transport.post("Nothing is running.", thread=msg.thread)
             return
+        if command in ("status", "details"):
+            # About the item, not a reply to it: nothing changes, a
+            # countdown runs on (L5). Recorded at the agent's level.
+            self._store.transcript(None, "operator").append(
+                UserMessage(text=text, source=f"transport:{msg.sender}")
+            )
+            said = (
+                self._details(item)
+                if command == "details"
+                else self._status_line(item, [i.id for i in self._queue()])
+            )
+            self._transport.post(said, thread=loop_of(item).thread)
+            return
         self._store.transcript(item.id, "operator").append(
             UserMessage(text=text, source=f"transport:{msg.sender}")
         )
@@ -844,8 +863,6 @@ class LoopAgent:
             self._close(item, loop, "abandoned", f"abandoned by {msg.sender}")
         elif command == "done" and loop.new_project:
             self._provisioned(item, loop)
-        elif command == "details":
-            self._transport.post(self._details(item), thread=loop.thread)
         elif (
             item.position == WAITING
             and item.scope != GENERAL
@@ -903,12 +920,13 @@ class LoopAgent:
             )
             self._go(item, loop, f"{msg.sender} (in plain language)")
         elif item.position == WAITING:
-            # An answer, or changes to a restatement: the driver reads it.
+            # An answer, changes to a restatement, or a remark: the driver
+            # reads it, and may reply and go back to this wait (L5).
             save(
                 self._store,
                 item.id,
                 READY,
-                replace(loop, wait="", note=""),
+                replace(loop, wait="", note="", back_to=loop.wait, back_note=loop.note),
                 cause="the operator answered",
             )
         self._schedule()
@@ -1272,28 +1290,28 @@ class LoopAgent:
                 f"{self._repos()}."
             )
         queue = [i.id for i in self._queue()]
-        lines = []
-        for item in items:
-            loop = loop_of(item)
-            what = item.position
-            if item.position == WAITING and loop.wait == CHAT:
-                what = "talking; your turn"
-            elif item.position == WAITING:
-                what = f"waiting for {loop.wait} ({loop.note or f'rev-{loop.rev}'})"
-                if loop.grace_until:
-                    left = max(0, int(loop.grace_until - self._clock()) // 60)
-                    what += f", going ahead in {left} min unless you reply"
-            elif item.position == BLOCKED:
-                what = f"blocked by {', '.join(loop.blocked_by)}"
-            elif item.position == READY:
-                what = f"ready, {queue.index(item.id) + 1} of {len(queue)} in line"
-            name = (
-                f'conversation "{item.title}"'
-                if item.scope == GENERAL
-                else f"{self._label(item)} {item.title}"
-            )
-            lines.append(f"{name}: {what}")
-        return "\n".join(lines)
+        return "\n".join(self._status_line(item, queue) for item in items)
+
+    def _status_line(self, item: WorkItem, queue: list[str]) -> str:
+        loop = loop_of(item)
+        what = item.position
+        if item.position == WAITING and loop.wait == CHAT:
+            what = "talking; your turn"
+        elif item.position == WAITING:
+            what = f"waiting for {loop.wait} ({loop.note or f'rev-{loop.rev}'})"
+            if loop.grace_until:
+                left = max(0, int(loop.grace_until - self._clock()) // 60)
+                what += f", going ahead in {left} min unless you reply"
+        elif item.position == BLOCKED:
+            what = f"blocked by {', '.join(loop.blocked_by)}"
+        elif item.position == READY:
+            what = f"ready, {queue.index(item.id) + 1} of {len(queue)} in line"
+        name = (
+            f'conversation "{item.title}"'
+            if item.scope == GENERAL
+            else f"{self._label(item)} {item.title}"
+        )
+        return f"{name}: {what}"
 
     # -- running (L1, L2) ------------------------------------------------------
 
@@ -1663,12 +1681,15 @@ class LoopAgent:
                     CLOSE_ITEM,
                     *([WAIT_FOR_REVIEW] if loop.pr else []),
                     *([DELEGATE, ASK_EXPERT] if self._helpers else []),
+                    *([REPLY] if loop.back_to else []),
                     *EXECUTOR_TOOLS,
                 ]
             ),
         ]
         if loop.question:
             tools = [REPLY, *_LOOK]
+        elif loop.back_to and not loop.started:
+            tools = [*tools, REPLY]  # before the go-ahead, a remark answered
         roles = self._roles_of(item)
         workspace = project.workspace
         system = SYSTEM.format(
@@ -1771,6 +1792,13 @@ class LoopAgent:
                 replace(loop, rev=rev, wait=GO_AHEAD, note="", grace_until=grace),
                 None,
             )
+        elif outcome.kind == "wait" and outcome.tool == "reply" and loop.back_to:
+            # A project item's reply: answered where asked (the thread or the
+            # PR), then back to the very wait the message interrupted (L5).
+            self._answer_where_asked(item, loop, str(args.get("text", "")).strip())
+            self._wait(
+                item, replace(loop, wait=loop.back_to, note=loop.back_note), None
+            )
         elif outcome.kind == "wait" and outcome.tool == "reply":
             text = str(args.get("text", "")).strip()
             self._transport.post(text, thread=loop.thread)
@@ -2068,6 +2096,8 @@ class LoopAgent:
     def _wait(self, item: WorkItem, loop: Loop, say: str | None) -> None:
         if say:
             self._transport.post(say, thread=loop.thread)
+        # Waiting again: whatever woke the item is settled.
+        loop = replace(loop, question=False, back_to="", back_note="")
         save(self._store, item.id, WAITING, loop, cause=f"waiting for {loop.wait}")
 
     def _close(self, item: WorkItem, loop: Loop, outcome: str, note: str) -> None:

nisse-dev/src/nisse_dev/loop/driver.py

@@ -12,7 +12,7 @@ from dataclasses import dataclass
 from typing import Any, Literal
 
 from nisse_dev.cancel import Cancel, skip_calls
-from nisse_dev.canonical import AssistantText, ToolCall, ToolResult
+from nisse_dev.canonical import AssistantText, Fact, ToolCall, ToolResult
 from nisse_dev.llm import LLMClient, LLMError
 from nisse_dev.loop.tools import WAITING, names
 from nisse_dev.state import Transcript
@@ -100,7 +100,10 @@ class Driver:
                 if nudged:
                     return RunOutcome("failed", note="kept replying in prose")
                 nudged = True
-                system = f"{system}\n\n{_PROSE_NUDGE}"
+                # In the conversation, as the harness: a request must end on
+                # the user's side (models refuse assistant prefill), and the
+                # nudge is then in the log as what it was.
+                transcript.as_actor("orchestrator").append(Fact(text=_PROSE_NUDGE))
                 continue
             for index, call in enumerate(calls):
                 if self._cancel.requested:

nisse-dev/src/nisse_dev/views.py

@@ -22,6 +22,7 @@ from dataclasses import replace
 from nisse_dev.canonical import (
     AssistantText,
     DelegationCompleted,
+    DelegationStarted,
     Fact,
     GateOpened,
     PlanSubmitted,
@@ -105,6 +106,13 @@ def project(events: list[StoredEvent], viewer: str) -> list[StoredEvent]:
             if viewer == "helper":
                 out.append(_said(stored, "approved plan", event.plan))
             continue
+        if isinstance(event, DelegationStarted):
+            # The harness handing a task over: to the helper given it, a
+            # message from the harness, never its own turn (a request must
+            # end on the user's side; models refuse assistant prefill).
+            if actor == viewer:
+                out.append(_said(stored, "orchestrator", f"your task: {event.brief}"))
+            continue
         if actor == viewer:
             out.append(stored)
         elif actor == "operator" and isinstance(event, UserMessage):