Merge pull request 'Fixes from the rest of the live session' (#122) from live-findings into main

Reviewed-on: #122
Reviewed-by: Don Garrett <dgarrett@acm.org>

.gitea/workflows/ci.yml

@@ -1,4 +1,10 @@
 name: CI
+# Three levels of testing:
+#   - every PR: lint, unit tests (16 property walks), integration tests, and
+#     the smoke end-to-end tests (`smoke` marker);
+#   - a PR labelled e2e:full, a push to main and a manual run: all 48 walks
+#     and the whole end-to-end suite;
+#   - a published release: the full set again before its image (release.yml).
 on:
   # Only main on push. A PR branch is already covered by the pull_request
   # event, and running both fired two identical suites concurrently — which
@@ -7,11 +13,18 @@ on:
   push:
     branches: [main]
   pull_request:
+    # labeled: adding e2e:full runs the full set; any other label, nothing.
+    types: [opened, synchronize, reopened, labeled]
   # Manual re-run, e.g. after rotating a credential or raising a budget cap.
   workflow_dispatch:
 
+env:
+  # Full on e2e:full, and outside PRs.
+  FULL: ${{ github.event_name != 'pull_request' || contains(github.event.pull_request.labels.*.name, 'e2e:full') }}
+
 jobs:
   checks:
+    if: github.event.action != 'labeled' || github.event.label.name == 'e2e:full'
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
@@ -26,12 +39,10 @@ jobs:
 
       - name: Install sshd
         # The tests reach "the build VM" over real SSH, through a throwaway
-        # sshd the test fixture starts (nisse_dev/testing.py).
-        run: |
-          SUDO=$(command -v sudo || true)
-          $SUDO apt-get update -qq
-          $SUDO apt-get install -y -qq openssh-server openssh-client > /dev/null
-          $SUDO mkdir -p /run/sshd
+        # sshd the test fixture starts (nisse_dev/testing.py). Skipped when the image has it; otherwise from
+        # a kernel.org mirror: archive.ubuntu.com has been slow enough to add
+        # minutes to every job.
+        run: scripts/ci-sshd.sh
 
       - name: Lint
         run: uv run ruff check .
@@ -40,7 +51,12 @@ jobs:
         run: uv run ruff format --check .
 
       - name: Unit tests
-        run: uv run pytest -m "not integration and not e2e"
+        # Across the runner's cores (pytest-xdist). On a PR, 16 property
+        # walks from a seed that moves with the run number; all 48 otherwise.
+        env:
+          NISSE_WALK_SEEDS: ${{ env.FULL == 'true' && '48' || '16' }}
+          NISSE_WALK_FIRST: ${{ env.FULL == 'true' && '0' || github.run_number }}
+        run: uv run pytest -n auto -m "not integration and not e2e"
 
       - name: Integration tests
         # Everything but the secrets is in tests/integration/nisse-test.yaml,
@@ -58,12 +74,15 @@ jobs:
   e2e:
     # The end-to-end suite (scripts/e2e.sh): the whole agent against real
     # services, every role on a local model, so it spends nothing but keeps
-    # the RTX 4090 busy for ~25 minutes. On PRs only, and only after the
-    # checks pass: a PR that's red already doesn't spend the GPU. There is
-    # one runner and Gitea doesn't honour cancel-in-progress, so a run a
-    # newer push supersedes holds the runner until it's cancelled by hand.
-    # Locally, run only what you're writing (-k) or diagnosing (--lf).
-    if: github.event_name != 'push'
+    # the RTX 4090 busy: ~10 minutes for the smoke tests, ~25+ for all. On
+    # PRs and manual runs, and only after the checks pass: a PR that's red
+    # already doesn't spend the GPU. There is one runner and Gitea doesn't
+    # honour cancel-in-progress, so a run a newer push supersedes holds the
+    # runner until it's cancelled by hand. Locally, run only what you're
+    # writing (-k) or diagnosing (--lf).
+    if: >-
+      github.event_name != 'push'
+      && (github.event.action != 'labeled' || github.event.label.name == 'e2e:full')
     needs: checks
     runs-on: ubuntu-latest
     timeout-minutes: 90
@@ -79,12 +98,10 @@ jobs:
         run: uv sync --all-packages
 
       - name: Install sshd
-        # The build VM is a throwaway sshd here (scripts/e2e.sh --local).
-        run: |
-          SUDO=$(command -v sudo || true)
-          $SUDO apt-get update -qq
-          $SUDO apt-get install -y -qq openssh-server openssh-client > /dev/null
-          $SUDO mkdir -p /run/sshd
+        # The build VM is a throwaway sshd here (scripts/e2e.sh --local). Skipped when the image has it; otherwise from
+        # a kernel.org mirror: archive.ubuntu.com has been slow enough to add
+        # minutes to every job.
+        run: scripts/ci-sshd.sh
 
       - name: End-to-end tests
         env:
@@ -93,4 +110,9 @@ jobs:
           NISSE_GITEA_TOKEN: ${{ secrets.NISSE_GITEA_TOKEN }}
           NISSE_OPERATOR_TOKEN: ${{ secrets.NISSE_OPERATOR_TOKEN }}
           NISSE_E2E_RUNS: /tmp/nisse-e2e-runs
-        run: scripts/e2e.sh --local -v
+        run: |
+          if [ "$FULL" = true ]; then
+            scripts/e2e.sh --local -v
+          else
+            scripts/e2e.sh --local -v -m "e2e and smoke"
+          fi

.gitea/workflows/release.yml

@@ -3,7 +3,8 @@ name: Release
 # registry (Development Plan M1.5, X4): gitea.dongar.net/dev/nisse-dev
 # tagged with the version (the release tag without its v) and latest. Cut a
 # release in Gitea (Releases → New release, tag vX.Y.Z); a tag pushed for any
-# other reason builds nothing.
+# other reason builds nothing. The full test set runs first: no image without
+# it passing.
 #
 # The registry needs a personal access token with write:package
 # (REGISTRY_TOKEN, an organization secret of `dev`, belonging to the
@@ -23,7 +24,46 @@ env:
   IMAGE: gitea.dongar.net/dev/nisse-dev
 
 jobs:
+  tests:
+    # Everything, before an image exists: lint, unit tests with all 48
+    # property walks, integration tests and the whole end-to-end suite.
+    runs-on: ubuntu-latest
+    timeout-minutes: 150
+    env:
+      NISSE_LLM_TOKEN: ${{ secrets.NISSE_LLM_TOKEN }}
+      NISSE_MATRIX_TOKEN: ${{ secrets.NISSE_MATRIX_TOKEN }}
+      NISSE_GITEA_TOKEN: ${{ secrets.NISSE_GITEA_TOKEN }}
+      NISSE_OPERATOR_TOKEN: ${{ secrets.NISSE_OPERATOR_TOKEN }}
+      NISSE_E2E_RUNS: /tmp/nisse-e2e-runs
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Install uv
+        run: |
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+          echo "$HOME/.local/bin" >> "$GITHUB_PATH"
+
+      - name: Sync workspace
+        run: uv sync --all-packages
+
+      - name: Install sshd
+        run: scripts/ci-sshd.sh
+
+      - name: Lint
+        run: uv run ruff check . && uv run ruff format --check .
+
+      - name: Unit tests
+        run: uv run pytest -n auto -m "not integration and not e2e"
+
+      - name: Integration tests
+        run: uv run pytest -m "integration and not e2e" -rs
+
+      - name: End-to-end tests
+        run: scripts/e2e.sh --local -v
+
   build-and-push:
+    needs: tests
     runs-on: ubuntu-latest
     permissions:
       contents: read

Dockerfile

@@ -10,6 +10,12 @@
 #   docker build -t nisse-dev .                  the agent
 #   docker build --target test -t nisse-dev:test .   plus the tests, for
 #                                                   scripts/e2e.sh --image
+#   docker build --target test --build-arg TEST_BASE=<published image> \
+#       -t nisse-dev:test .                      the tests over a release as
+#                                                published (M1.5 acceptance)
+
+# What the test stage stacks on: this build's agent, or a published image.
+ARG TEST_BASE=runtime
 
 FROM python:3.12-slim AS build
 COPY --from=ghcr.io/astral-sh/uv:0.12.10 /uv /usr/local/bin/uv
@@ -40,9 +46,10 @@ VOLUME /var/lib/nisse-dev
 USER nisse
 ENTRYPOINT ["nisse-dev"]
 
-# The image's own package plus the tests, so the end-to-end suite runs
-# against what was built, not the source tree.
-FROM runtime AS test
+# The tests over an image's own package (this build's, or a published one,
+# TEST_BASE): only the test tools are added, never the agent reinstalled, so
+# the end-to-end suite runs against what was built, not the source beside it.
+FROM ${TEST_BASE} AS test
 USER root
 RUN apt-get update \
     && apt-get install -y --no-install-recommends git openssh-server \
@@ -57,11 +64,9 @@ COPY nisse-ops nisse-ops
 COPY tests tests
 COPY scripts scripts
 COPY conftest.py ./
-ENV UV_PROJECT_ENVIRONMENT=/opt/nisse
-# Test tools too, all installed (not editable), and `uv run` never re-syncs:
-# the tests import the package as built, not the source tree beside them.
-RUN uv sync --frozen --all-packages --no-editable --inexact
-ENV UV_NO_SYNC=1
+ENV UV_PROJECT_ENVIRONMENT=/opt/nisse UV_NO_SYNC=1
+RUN uv export --frozen --only-group dev --no-hashes --no-emit-project > /tmp/dev.txt \
+    && uv pip install --quiet --python /opt/nisse/bin/python -r /tmp/dev.txt
 ENTRYPOINT []
 
 # The default target: the agent itself.

docs/Architecture.md

@@ -569,6 +569,9 @@ Work arrives as Gitea issues or conversation, and leaves as reviewed PRs.
   is. The agent acknowledges it ("queued myproject#14, 1 ahead"). Your issue is the
   item's record. Only issues filed or assigned by allowlisted operators count (L11).
 - **Unassigned issues are backlog,** never work, though the agent will discuss them.
+- **An issue assigned again after its item closed is a new item,** which notes the
+  one before it; an issue whose item is still open, or whose assignment predates
+  the close, stays with that item.
 - **Open Gitea dependencies make an issue blocked** (L1).
 - **Work that starts in conversation gets a record** once it is work on a project,
   not at its first message: the transcript already keeps the conversation. The

docs/Development-Plan.md

@@ -73,10 +73,14 @@ Only `nisse-dev` has real code; `nisse-core` and `nisse-ops` are placeholders.
   `CLAUDE.md`. A milestone marked **needs discussion** starts with a design
   discussion first.
 - **Every change comes with tests,** through `scripts/check.sh`.
-- **The end-to-end test runs on every PR** (M1.6): the real agent, every role on a
-  local model (so it costs nothing per call), a repository of the run's own and a
-  real build host, driven by a scripted operator. CI runs it as its own job on pull
-  requests, after the checks pass. There is one runner, and a run a newer push
+- **Testing has three levels** (M1.6): every PR runs the checks (unit tests across
+  the runner's cores, 16 property walks from a moving seed, integration) and the
+  end-to-end **smoke** tests; a PR labelled `e2e:full`, a push to `main` and a
+  manual run add all 48 walks and the whole end-to-end suite; a published release
+  runs that full set before its image is built. The end-to-end tests drive the
+  real agent, every role on a local model (so they cost nothing per call), in a
+  repository of the run's own, with a scripted operator; CI runs them after the
+  checks pass. There is one runner, and a run a newer push
   supersedes has to be cancelled by hand. Locally only the tests being written
   (`-k`) or diagnosed (`--lf`); `scripts/e2e.sh` reports what a run spent. Manual
   live runs are for the new loop's first runs and phase ends. Findings are filed as
@@ -271,8 +275,9 @@ in prose, asking mid-work (L5, #42, #77).
 - Deferred: config reload without a restart (with M2.5's entry step if wanted);
   the trust dial's configuration (M3.3).
 
-**M1.5 The agent as a container image.** *Pre-approved.* **Built** (this PR);
-accepted once a tag's image passes the e2e suite.
+**M1.5 The agent as a container image.** *Pre-approved.* **Done** (#115; accepted
+2026-10-07: `v0.1.0`'s published image passed the whole e2e suite, 9 of 9, run with
+the release's own tests over it).
 - A Dockerfile for `nisse-dev`, stateless with a volume for state; a Gitea Actions
   workflow that builds it and pushes to Gitea's registry on each release tag (X4).
 - `bringup.sh` runs the image instead of a venv.
@@ -283,6 +288,10 @@ accepted once a tag's image passes the e2e suite.
   package, for `scripts/e2e.sh --image REF`. Images are built only on release:
   `release.yml` pushes `gitea.dongar.net/dev/nisse-dev:<version>` and `:latest`
   on a published release (`REGISTRY_TOKEN`, the `dev` organization's).
+- Acceptance runs the tests over a published image: `docker build --target test
+  --build-arg TEST_BASE=gitea.dongar.net/dev/nisse-dev:<version>` from the
+  release's checkout, then `scripts/e2e.sh --image`; the test stage adds only the
+  test tools, so the agent under test is the one published.
 - The live bring-up's switch to the image waits for the next live test (it needs
   its own test repository too); `bringup.sh plan` shows what it will run.
 

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/gitea.py

@@ -261,6 +261,31 @@ class GiteaClient:
                 who = (event.get("user") or {}).get("login")
         return who
 
+    def assigned_at(self, repo: str, index: int, assignee: str) -> str | None:
+        """When the issue was last assigned to ``assignee`` (ISO 8601), from
+        its timeline, oldest first and paged to the end; None if the
+        timeline doesn't say."""
+        when = None
+        for page in range(1, MAX_PR_PAGES + 1):
+            events = (
+                self._request(
+                    "GET",
+                    f"/repos/{repo}/issues/{index}/timeline",
+                    params={"limit": PR_PAGE_SIZE, "page": page},
+                )
+                or []
+            )
+            for event in events:
+                if (
+                    event.get("type") == "assignees"
+                    and not event.get("removed_assignee")
+                    and (event.get("assignee") or {}).get("login") == assignee
+                ):
+                    when = event.get("created_at")
+            if len(events) < PR_PAGE_SIZE:
+                break
+        return when
+
     def repository(self, repo: str) -> dict[str, Any]:
         """The repository as this account sees it, permissions included."""
         return dict(self._request("GET", f"/repos/{repo}"))

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

@@ -22,6 +22,7 @@ import time
 import uuid
 from collections.abc import Callable, Iterable, Sequence
 from dataclasses import dataclass, replace
+from datetime import UTC, datetime
 
 from nisse_dev.cancel import Cancel
 from nisse_dev.canonical import (
@@ -87,6 +88,19 @@ _NEXT = re.compile(rf"(?:do|work on) {_ISSUE} (?:next|first|now)")
 _NAMED = re.compile(r"(go|approve|abandon|stop) (?:([a-z0-9][a-z0-9-]*))?#(\d+)")
 # At top level these need an item; without one they are ambiguous (L5).
 _NEEDS_ITEM = frozenset({"go", "approve", "abandon"})
+# Near-misses for abandon: asked about, never taken for it or for a reply.
+_NEAR_ABANDON = frozenset(
+    {
+        "abort",
+        "abort it",
+        "cancel",
+        "cancel it",
+        "drop it",
+        "quit",
+        "kill it",
+        "scrap it",
+    }
+)
 BUG_GRACE_S = 600.0
 GENERAL_IDLE_S = 86400.0
 PROGRESS_EVERY_S = 60.0  # progress in prose at a readable rate (L5, #42)
@@ -113,9 +127,16 @@ If the request is unclear, ask_human instead of guessing: a restatement asks
 no questions, so the operator can agree to it as it stands. At any point,
 asking is better than guessing.
 
+Only the operator stops or closes an item (their stop and abandon commands):
+never say you have stopped or closed anything yourself.
+
 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.
 
@@ -500,7 +521,8 @@ class LoopAgent:
             self._oplog.emit("loop.intake_unavailable", cause=str(exc))
             return
         for issue in issues:
-            if item_for_issue(self._store, issue.number, project.name) is not None:
+            before = item_for_issue(self._store, issue.number, project.name)
+            if before is not None and not self._assigned_again(project, issue, before):
                 continue
             if _MARKER.search(issue.body or ""):
                 continue  # D1: a record of work started in conversation
@@ -517,7 +539,7 @@ class LoopAgent:
             except GiteaError as exc:
                 self._oplog.emit("loop.intake_unavailable", cause=str(exc))
                 continue  # taken on the next round
-            self._open(project, issue, tuple(blocked_by))
+            self._open(project, issue, tuple(blocked_by), follows=before)
 
     # -- PR review and merge (D4, D5) -------------------------------------------
 
@@ -587,6 +609,8 @@ 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="")
@@ -683,8 +707,24 @@ class LoopAgent:
             return False
         return who in self._gitea_operators
 
+    def _assigned_again(self, project: Project, issue: Issue, before: WorkItem) -> bool:
+        """D1: an issue is taken once, and again only when it's assigned to
+        the agent anew after its last item closed (abandoned, say, then
+        given back): never on the same assignment twice."""
+        if before.status == "open" or not before.closed_at:
+            return False
+        try:
+            when = self._gitea.assigned_at(project.repo, issue.number, self._account)
+        except GiteaError:
+            return False
+        return bool(when) and _instant(when) > _instant(before.closed_at)
+
     def _open(
-        self, project: Project, issue: Issue, blocked_by: tuple[str, ...]
+        self,
+        project: Project,
+        issue: Issue,
+        blocked_by: tuple[str, ...],
+        follows: WorkItem | None = None,
     ) -> None:
         item_id = f"{project.name}-{issue.number}-{uuid.uuid4().hex[:6]}"
         cause = f"assigned {project.repo}#{issue.number}"
@@ -702,6 +742,7 @@ class LoopAgent:
             issue=issue.number,
             blocked_by=blocked_by,
             labels=labels,
+            follows=follows.id if follows else "",
         )
         self._threads[(project.name, issue.number)] = root
         self._store.open_work_item(
@@ -723,6 +764,12 @@ class LoopAgent:
                     if project.labels
                     else ""
                 )
+                + (
+                    f"; assigned again after its earlier item {follows.id} closed "
+                    f"({follows.outcome})"
+                    if follows
+                    else ""
+                )
             )
         )
         text = f"{issue.title}\n\n{issue.body}".strip()
@@ -824,6 +871,31 @@ class LoopAgent:
             elif not self._store.active_item():
                 self._transport.post("Nothing is running.", thread=msg.thread)
             return
+        if command in _NEAR_ABANDON:
+            # L5: a near-miss for a command is asked about, never read as a
+            # reply (the driver can't stop or close anything itself).
+            self._store.transcript(None, "operator").append(
+                UserMessage(text=text, source=f"transport:{msg.sender}")
+            )
+            self._transport.post(
+                f"Did you mean `abandon`? That closes {self._label(item)} and its "
+                "issue; say abandon to confirm. (`stop` ends a run in progress.)",
+                thread=loop_of(item).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 +916,6 @@ 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 +973,13 @@ 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 +1343,28 @@ 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) ------------------------------------------------------
 
@@ -1648,7 +1719,9 @@ class LoopAgent:
             loop = replace(loop, prepared=True)
         save(self._store, item.id, RUNNING, loop, cause="driver run")
         # D1: a `discuss` item talks before it restates.
-        hold = "discuss" in loop.labels and not loop.talked and not loop.started
+        hold = (
+            "discuss" in loop.labels and not loop.talked
+        ) or loop.started  # agreed: a restatement is behind it (D1)
         if loop.question:
             hold = True  # nothing but answering (below)
         tools = [
@@ -1663,12 +1736,15 @@ 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 +1847,13 @@ 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 +2151,8 @@ 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:
@@ -2094,8 +2179,19 @@ class LoopAgent:
                 self._gitea.post_comment(project.repo, loop.issue, f"{outcome}: {note}")
                 self._gitea.close_issue(project.repo, loop.issue)
             except GiteaError as exc:
+                try:
+                    blockers = self._gitea.open_dependencies(project.repo, loop.issue)
+                except GiteaError:
+                    blockers = []
+                why = (
+                    f"Gitea won't close an issue while it depends on open ones "
+                    f"({', '.join(blockers)}): close it when they're done, or "
+                    "remove the dependency"
+                    if blockers
+                    else str(exc)
+                )
                 self._transport.post(
-                    f"closed here, but could not close #{loop.issue}: {exc}",
+                    f"closed here, but could not close #{loop.issue}: {why}",
                     thread=loop.thread,
                 )
         if outcome == "abandoned":
@@ -2103,3 +2199,9 @@ class LoopAgent:
 
     def _fact(self, item: WorkItem, text: str) -> None:
         self._store.transcript(item.id, "orchestrator").append(Fact(text=text))
+
+
+def _instant(stamp: str) -> datetime:
+    """An ISO 8601 time from Gitea or the state store, comparable either way."""
+    when = datetime.fromisoformat(stamp)
+    return when if when.tzinfo else when.replace(tzinfo=UTC)

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/loop/items.py

@@ -108,9 +108,12 @@ def item_for_thread(
 def item_for_issue(
     store: StateStore, issue: int, project: str | None = None
 ) -> WorkItem | None:
-    """Any item, open or closed, recorded by this issue (in ``project``,
-    when given): intake never takes an issue twice (D1)."""
-    for item in store.work_items():
-        if loop_of(item).issue == issue and project in (None, item.scope):
-            return item
-    return None
+    """The latest item, open or closed, recorded by this issue (in
+    ``project``, when given): intake takes an issue once, and again only on a
+    fresh assignment after its item closed (D1)."""
+    found = [
+        item
+        for item in store.work_items()
+        if loop_of(item).issue == issue and project in (None, item.scope)
+    ]
+    return found[-1] if found else None

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):

nisse-dev/tests/model_contract.py

@@ -20,6 +20,13 @@ def assert_renderable(events: list[Any]) -> None:
 
 
 def _anthropic(messages: list[dict[str, Any]]) -> None:
+    # A request ends on the user's side: current models refuse assistant
+    # prefill ("The conversation must end with a user message"), found live
+    # on 2026-10-07 when the driver's prose nudge left its own reply last.
+    assert not messages or messages[-1]["role"] == "user", (
+        f"anthropic messages.{len(messages) - 1}: the conversation ends with an "
+        "assistant message (prefill), which the API refuses"
+    )
     for index, message in enumerate(messages):
         if message["role"] != "assistant":
             continue

nisse-dev/tests/test_gitea.py

@@ -229,3 +229,30 @@ def test_closing_comments_first_then_closes_issue_and_pr() -> None:
         ("POST", "/api/v1/repos/o/r/issues/20/comments", {"body": "abandoned by don"}),
         ("PATCH", "/api/v1/repos/o/r/pulls/20", {"state": "closed"}),
     ]
+
+
+def test_assigned_at_is_the_last_assignment_across_pages() -> None:
+    def assigned(login: str, at: str, removed: bool = False) -> dict:
+        return {
+            "type": "assignees",
+            "assignee": {"login": login},
+            "removed_assignee": removed,
+            "created_at": at,
+        }
+
+    filler = {"type": "comment", "created_at": "t0"}
+    pages = {
+        1: [assigned("dev-agent", "t1")] + [filler] * (PR_PAGE_SIZE - 1),
+        2: [
+            assigned("dev-agent", "t2", removed=True),
+            assigned("someone", "t3"),
+            assigned("dev-agent", "t4"),
+        ],
+    }
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        assert request.url.path == "/api/v1/repos/o/r/issues/6/timeline"
+        return httpx.Response(200, json=pages[int(request.url.params["page"])])
+
+    assert client(handler).assigned_at("o/r", 6, "dev-agent") == "t4"
+    assert client(handler).assigned_at("o/r", 6, "nobody") is None

nisse-dev/tests/test_loop.py

@@ -49,6 +49,7 @@ class FakeGitea:
         self.comments: list[tuple[int, str]] = []
         self.closed: list[int] = []
         self.deps: dict[int, list[int]] = {}
+        self.reassigned: dict[int, str] = {}  # issue -> when assigned again
 
     def create_issue(self, repo: str, title: str, body: str) -> int:
         n = max([i.number for i in self.issues] + [0]) + 1
@@ -67,6 +68,9 @@ class FakeGitea:
     def assigned_by(self, repo: str, index: int, assignee: str) -> str:
         return self.assigner
 
+    def assigned_at(self, repo: str, index: int, assignee: str) -> str | None:
+        return self.reassigned.get(index)
+
     def post_comment(self, repo: str, index: int, body: str) -> None:
         self.comments.append((index, body))
 

nisse-dev/tests/test_loop_live_fixes.py

@@ -0,0 +1,238 @@
+"""What the first live session of the new loop found (2026-10-07), each as a
+test: a request ending on the assistant's side (refused by models without
+prefill: the driver's prose nudge, a helper's first request), `status` in an
+item's thread taken as a reply, no way for the driver to answer a remark;
+later, near-miss commands, the driver claiming a stop, a reassigned issue
+ignored, an unexplained failed abandon and restating an agreed item.
+"""
+
+from pathlib import Path
+from typing import Any
+
+from model_contract import assert_renderable
+from test_loop import DON, RESTATED, FakeGitea, Harness, ScriptedModel, call, issue
+from test_loop_help import ROLES, hand_back
+from test_loop_milestones import RESTATE, Forge, loop_agent, say
+from test_loop_milestones import item as the_item
+from test_loop_milestones import world as world  # noqa: PLC0414 -- the fixture
+
+from nisse_dev.canonical import AssistantText, Fact
+from nisse_dev.delegation import Caps
+from nisse_dev.gitea import Comment
+from nisse_dev.loop.help import Helpers
+from nisse_dev.loop.items import GO_AHEAD, MERGE, WAITING, loop_of
+
+
+class Clock:
+    def __init__(self) -> None:
+        self.now = 1000.0
+
+    def __call__(self) -> float:
+        return self.now
+
+
+def test_the_prose_nudge_keeps_requests_ending_on_the_user_side(
+    tmp_path: Path,
+) -> None:
+    # ScriptedModel checks every request as the API would, now including
+    # "the conversation must end with a user message".
+    h = Harness(
+        tmp_path,
+        ScriptedModel([AssistantText(text="Let me think.", model="m")], RESTATED),
+        FakeGitea(issue()),
+    )
+    h.agent.intake()
+    assert loop_of(h.item()).wait == GO_AHEAD  # the nudge worked
+    facts = [
+        e.event.text
+        for e in h.store.transcript(h.item().id, "orchestrator").read()
+        if isinstance(e.event, Fact)
+    ]
+    assert any("Your last reply was prose" in f for f in facts)
+
+
+class ContractHelper:
+    """A helper that checks its requests as the API would."""
+
+    def __init__(self, *turns: list[Any]) -> None:
+        self.turns = list(turns)
+
+    def complete(self, events, tools, model, system=None) -> list[Any]:
+        assert_renderable(events)
+        return self.turns.pop(0)
+
+
+def test_a_helpers_requests_end_on_the_user_side(world: dict[str, Any]) -> None:
+    helper = ContractHelper(
+        [AssistantText(text="Looking around first.", model="h")],  # prose
+        [call("write_file", path="helped.py", content="H = 1\n")],
+        [hand_back()],
+    )
+    model = ScriptedModel(
+        RESTATE,
+        [
+            call("start_milestone", name="words"),
+            call("delegate", task="write helped.py"),
+            call("finish_milestone", summary="with help", final="yes"),
+        ],
+    )
+    helpers = Helpers(
+        client=helper,  # type: ignore[arg-type]
+        roles=ROLES,
+        client_for=lambda model: helper,  # type: ignore[arg-type,return-value]
+        caps=Caps(turns=5),
+    )
+    agent, transport = loop_agent(world, model, helpers=helpers)
+    agent.intake()
+    say(transport, world, "go")
+    forge: Forge = world["forge"]
+    [pr] = forge.prs
+    assert forge.files(pr) == ["helped.py"]
+
+
+def test_status_in_a_thread_is_not_a_reply(tmp_path: Path) -> None:
+    clock = Clock()
+    model = ScriptedModel(RESTATED)  # called once only: status runs nothing
+    h = Harness(
+        tmp_path,
+        model,
+        FakeGitea(issue(labels=("bug",))),
+        clock=clock,
+        bug_grace_s=600,
+    )
+    h.agent.intake()
+    grace = loop_of(h.item()).grace_until
+    h.say("status")
+    assert h.posted()[-1].startswith("#14 Add count_words: waiting for go-ahead")
+    loop = loop_of(h.item())
+    assert (h.item().position, loop.wait, loop.grace_until) == (
+        WAITING,
+        GO_AHEAD,
+        grace,
+    )
+    h.say("details")
+    assert loop_of(h.item()).grace_until == grace
+
+
+def test_a_remark_gets_a_reply_and_the_restatement_stays_up(tmp_path: Path) -> None:
+    h = Harness(
+        tmp_path,
+        ScriptedModel(
+            RESTATED,
+            [call("reply", text="Noted: I'll keep it to text.py.")],
+        ),
+        FakeGitea(issue()),
+    )
+    h.agent.intake()
+    h.say("just a thought: keep it small")
+    loop = loop_of(h.item())
+    assert (loop.wait, loop.rev) == (GO_AHEAD, 1)
+    assert h.posted()[-1] == "Noted: I'll keep it to text.py."
+
+
+def test_a_review_comment_can_be_answered_without_a_fix(
+    world: dict[str, Any],
+) -> None:
+    forge: Forge = world["forge"]
+    model = ScriptedModel(
+        RESTATE,
+        [
+            call("start_milestone", name="words"),
+            call("write_file", path="w.py", content="W = 1\n"),
+            call("finish_milestone", summary="words", final="yes"),
+        ],
+        [call("reply", text="It's a constant on purpose, for the test.")],
+    )
+    agent, transport = loop_agent(world, model)
+    agent.intake()
+    say(transport, world, "go")
+    [pr] = forge.prs
+    forge.comments[pr] = [Comment(7, "don", "why a constant", "t")]
+    agent.watch()
+    agent._schedule()
+    loop = loop_of(the_item(world))
+    assert (loop.wait, loop.pr) == (MERGE, pr)
+    assert forge.posted[-1][0] == pr  # answered on the PR, where asked
+    assert "It's a constant on purpose" in forge.posted[-1][1]
+
+
+# --- the session's later findings ----------------------------------------------
+
+
+def test_a_near_miss_for_abandon_is_asked_about(tmp_path: Path) -> None:
+    model = ScriptedModel(RESTATED)  # called once only: nothing reaches it
+    h = Harness(tmp_path, model, FakeGitea(issue()))
+    h.agent.intake()
+    h.say("abort")
+    assert h.posted()[-1].startswith("Did you mean `abandon`?")
+    assert (h.item().status, loop_of(h.item()).wait) == ("open", GO_AHEAD)
+    h.say("abandon")
+    assert h.item().status == "closed"
+
+
+def test_the_driver_is_told_it_cant_stop_or_close(tmp_path: Path) -> None:
+    seen: list[str] = []
+
+    class Spy(ScriptedModel):
+        def complete(self, events, tools, model, system=None):  # type: ignore[no-untyped-def]
+            seen.append(system or "")
+            return super().complete(events, tools, model, system)
+
+    h = Harness(tmp_path, Spy(RESTATED), FakeGitea(issue()))
+    h.agent.intake()
+    assert "never say you have stopped or closed anything yourself" in seen[0]
+
+
+def test_an_issue_assigned_again_after_its_item_closed_is_taken_again(
+    tmp_path: Path,
+) -> None:
+    gitea = FakeGitea(issue())
+    h = Harness(tmp_path, ScriptedModel(RESTATED, RESTATED), gitea)
+    h.agent.intake()
+    h.say("abandon")
+    [old] = h.store.work_items()
+    gitea.closed.remove(14)  # the operator reopened it
+    h.agent.intake()
+    assert len(h.store.work_items()) == 1  # the same assignment: not again
+    gitea.reassigned[14] = "2999-01-01T00:00:00Z"  # assigned anew, later
+    h.agent.intake()
+    items = h.store.work_items()
+    assert len(items) == 2
+    new = next(i for i in items if i.id != old.id)
+    assert new.status == "open" and loop_of(new).follows == old.id
+
+
+def test_abandoning_a_blocked_issue_says_why_it_stays_open(tmp_path: Path) -> None:
+    from nisse_dev.gitea import GiteaError
+
+    class Strict(FakeGitea):
+        def close_issue(self, repo: str, index: int) -> None:
+            if self.open_dependencies(repo, index):
+                raise GiteaError("gitea PATCH returned 500: blocked")
+            super().close_issue(repo, index)
+
+    gitea = Strict(issue(14), issue(15, title="other"))
+    gitea.deps[14] = [15]
+    h = Harness(tmp_path, ScriptedModel(RESTATED), gitea)
+    h.agent._running = True  # hold the queue: only the abandon matters
+    h.agent.intake()
+    item = next(i for i in h.store.work_items() if loop_of(i).issue == 14)
+    h.transport.receive(DON, "abandon", thread=loop_of(item).thread)
+    said = " ".join(h.posted())
+    assert "won't close an issue while it depends on open ones" in said
+    assert "don/sandbox#15" in said
+
+
+def test_an_agreed_item_isnt_offered_restate(world: dict[str, Any]) -> None:
+    offered: list[set[str]] = []
+
+    class Spy(ScriptedModel):
+        def complete(self, events, tools, model, system=None):  # type: ignore[no-untyped-def]
+            offered.append({t["function"]["name"] for t in tools})
+            return super().complete(events, tools, model, system)
+
+    model = Spy(RESTATE, [call("ask_human", question="anything else?")])
+    agent, transport = loop_agent(world, model)
+    agent.intake()
+    say(transport, world, "go")
+    assert "restate" in offered[0] and "restate" not in offered[1]

nisse-dev/tests/test_loop_milestones.py

@@ -61,6 +61,7 @@ class Forge:
         self.issues = [Issue(14, "Add count_words", "count words", "don", ())]
         self.deps: dict[int, list[int]] = {}  # issue -> the issues it depends on
         self.assigned: list[int] = []  # issues the agent assigned to itself
+        self.reassigned: dict[int, str] = {}  # issue -> when assigned again
 
     # intake
     def assigned_issues(self, repo: str, assignee: str) -> list[Issue]:
@@ -84,6 +85,9 @@ class Forge:
     def assigned_by(self, repo: str, index: int, assignee: str) -> str:
         return "don"
 
+    def assigned_at(self, repo: str, index: int, assignee: str) -> str | None:
+        return self.reassigned.get(index)
+
     # PRs
     def _pr(self, n: int) -> PullRequest:
         p = self.prs[n]

nisse-dev/tests/test_loop_properties.py

@@ -8,6 +8,7 @@ checks every property after every step. A failure prints
 the seed and the steps, so it replays exactly.
 """
 
+import os
 import random
 import re
 import subprocess
@@ -46,8 +47,14 @@ from nisse_dev.tools import Workspace
 from nisse_dev.transport import InboundMessage, InMemoryTransport
 from nisse_dev.worker import InlineWorker
 
-SEEDS = range(32)
-TWO_PROJECT_SEEDS = range(100, 116)
+# How many walks, and from which seed: all 48 by default; CI runs fewer on a
+# PR, from a seed that moves with the run number, so PRs still cover new
+# ground (NISSE_WALK_SEEDS, NISSE_WALK_FIRST). Two thirds walk one project,
+# a third two.
+_WALKS = int(os.environ.get("NISSE_WALK_SEEDS") or 48)
+_FIRST = int(os.environ.get("NISSE_WALK_FIRST") or 0)
+SEEDS = range(_FIRST, _FIRST + _WALKS * 2 // 3)
+TWO_PROJECT_SEEDS = range(100 + _FIRST, 100 + _FIRST + _WALKS - _WALKS * 2 // 3)
 STEPS = 30
 
 
@@ -76,8 +83,15 @@ class PolicyModel:
             return [call("ask_human", question=f"question {self.n}?")]
         if roll < 0.12:
             return [ToolCall(call_id=f"p{self.n}", tool="nonsense", arguments={})]
-        if "reply" in offered and "propose_work" not in offered:
+        if "reply" in offered and not offered & {
+            "propose_work",
+            "restate",
+            "ask_human",
+            "start_milestone",
+        }:
             return [call("reply", text=f"because {self.n}")]  # a question run
+        if "reply" in offered and "propose_work" not in offered and roll < 0.2:
+            return [call("reply", text=f"noted {self.n}")]  # a remark answered
         if "reply" in offered:  # a conversation: talk, or propose work
             if self.rng.random() < 0.1:
                 return [
@@ -251,6 +265,7 @@ class Walk:
         if item and loop:
             weights.update({"go": 3, "answer": 2, "abandon": 1, "go-stale": 1})
             weights.update({"plain-yes": 2, "wait": 2, "question": 2})
+            weights.update({"status-here": 1})
             if loop.new_project:
                 weights.update({"done": 2})
             if item.scope != "general":  # conversations aren't named by number
@@ -266,6 +281,18 @@ class Walk:
         elif action == "done" and loop:
             # The operator says a new project is ready; no entry was added.
             self.transport.receive(DON, "done", thread=loop.thread)
+        elif action == "status-here" and loop:
+            # L5: a command about the item, not a reply: nothing changes,
+            # not even a countdown.
+            before = (item.position, loop.wait, loop.rev, loop.grace_until)
+            self.transport.receive(DON, "status", thread=loop.thread)
+            after_item = self.world["store"].work_item(item.id)
+            after = loop_of(after_item)
+            now = (after_item.position, after.wait, after.rev, after.grace_until)
+            assert now == before, (
+                f"status_changes_nothing: {before} -> {now}: "
+                f"seed {self.seed}, steps {self.steps}"
+            )
         elif action == "question" and loop:
             # L5: a question never changes the item it's asked about.
             before = (item.position, loop.wait, loop.rev, loop.pr, loop.branch)
@@ -373,7 +400,13 @@ class Walk:
         where = f"seed {self.seed}, steps {self.steps}"
         store = self.world["store"]
         items = self.items()
-        issues = [(i.scope, loop_of(i).issue) for i in items if loop_of(i).issue]
+        # Open items: an issue assigned again after its item closed gets a new
+        # one (D1), which loop.qnt doesn't model.
+        issues = [
+            (i.scope, loop_of(i).issue)
+            for i in items
+            if loop_of(i).issue and i.status == "open"
+        ]
         assert len(issues) == len(set(issues)), f"one_item_per_issue: {where}"
         for item in items:
             loop = loop_of(item)

pyproject.toml

@@ -4,6 +4,7 @@ members = ["nisse-core", "nisse-dev", "nisse-ops"]
 [dependency-groups]
 dev = [
     "pytest>=8",
+    "pytest-xdist>=3",
     "ruff>=0.6",
 ]
 
@@ -19,6 +20,7 @@ addopts = "--strict-markers --ignore=tests/integration/seed"
 markers = [
     "integration: end-to-end tests that exercise installed entry points or external services",
     "e2e: the whole agent against real services, run after each milestone (scripts/e2e.sh)",
+    "smoke: the end-to-end tests every PR runs; the rest on e2e:full and before a release",
 ]
 
 [tool.ruff]

scripts/bringup-repo.sh

@@ -0,0 +1,42 @@
+#!/usr/bin/env bash
+# The bring-up's test repository (Development Plan, live runs):
+# nisse-testing-operator/nisse-bringup, made as the end-to-end tests make
+# theirs (tests/integration/e2e_support.py, make_repo), but kept between runs.
+# Seeded with the test package, the bring-up agent's account a collaborator,
+# main protected, and a deploy key of its own at $E2E/ssh/nisse-bringup.key.
+#
+#   scripts/bringup-repo.sh create    make it
+#   scripts/bringup-repo.sh reset     delete it and make it again
+#   scripts/bringup-repo.sh delete    delete it
+#
+# Needs NISSE_OPERATOR_TOKEN (.env here) and the bring-up agent's token
+# (NISSE_GITEA_TOKEN in $E2E/.env), which names the agent's account.
+set -euo pipefail
+cd "$(dirname "$0")/.."
+E2E=${NISSE_E2E:-/root/nisse-e2e}
+case "${1:-}" in create|reset|delete) ;; *) sed -n '2,15p' "$0" | sed 's/^# \{0,1\}//'; exit 2 ;; esac
+set -a; [ -f .env ] && . ./.env; set +a
+exec uv run python - "$1" "$E2E" <<'PY'
+import sys
+from pathlib import Path
+
+sys.path.insert(0, "tests/integration")
+from e2e_support import OWNER, delete_repo, make_repo, operator_gitea  # noqa: E402
+
+from nisse_dev.envfile import load_env_file  # noqa: E402
+from nisse_dev.gitea import GiteaClient  # noqa: E402
+
+action, e2e = sys.argv[1], Path(sys.argv[2])
+url = "https://gitea.dongar.net"
+operator = operator_gitea(url)
+name = "nisse-bringup"
+if action in ("reset", "delete"):
+    delete_repo(operator, f"{OWNER}/{name}")
+    print(f"deleted {OWNER}/{name}")
+if action in ("create", "reset"):
+    agent_env: dict[str, str] = {}
+    load_env_file(e2e / ".env", agent_env)
+    agent = GiteaClient(url, agent_env["NISSE_GITEA_TOKEN"]).whoami()
+    repo = make_repo(operator, name, e2e / "ssh", agent=agent)
+    print(f"made {repo.name}: {agent} a collaborator; push key {repo.push_key}")
+PY

scripts/ci-sshd.sh

@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+# The throwaway sshd the tests need (nisse_dev/testing.py), in a CI job.
+# Skipped when the job's image already has it. Otherwise apt fetches from a
+# kernel.org mirror rather than archive.ubuntu.com, which has been slow
+# enough (a few KB/s) to add minutes to every job.
+set -euo pipefail
+SUDO=$(command -v sudo || true)
+if command -v /usr/sbin/sshd > /dev/null && command -v ssh > /dev/null; then
+    echo "sshd already installed"
+else
+    MIRROR=${APT_MIRROR:-http://mirrors.edge.kernel.org/ubuntu}
+    # Both layouts: deb822 (ubuntu.sources, 24.04+) and sources.list (older).
+    for f in /etc/apt/sources.list.d/ubuntu.sources /etc/apt/sources.list; do
+        [ -f "$f" ] && $SUDO sed -i -E \
+            "s#https?://(archive|security)\.ubuntu\.com/ubuntu/?#$MIRROR/#g" "$f"
+    done
+    $SUDO apt-get update -qq
+    $SUDO apt-get install -y -qq openssh-server openssh-client > /dev/null
+fi
+$SUDO mkdir -p /run/sshd

scripts/e2e.sh

@@ -23,6 +23,8 @@ if [ "${1:-}" = "--image" ]; then
     mkdir -p "$runs"
     secrets=()
     [ -f .env ] && secrets=(--env-file .env)
+    # The model, if overridden here, reaches the tests inside too.
+    [ -n "${NISSE_E2E_MODEL:-}" ] && secrets+=(-e "NISSE_E2E_MODEL=$NISSE_E2E_MODEL")
     # Host networking: the suite reaches Matrix, Gitea and LiteLLM as here.
     exec docker run --rm --network host "${secrets[@]}" \
         -e NISSE_E2E_RUNS=/runs -v "$runs:/runs" \

tests/integration/e2e_support.py

@@ -15,8 +15,10 @@ from nisse_dev.llm import RoleMap
 
 # The end-to-end tests run every role on a local model, which costs nothing
 # per call; NISSE_E2E_MODEL overrides it (e.g. claude-haiku-4-5 for a quick,
-# paid run).
-E2E_MODEL = "qwen3.8-27b"
+# paid run). The no-thinking alias (LiteLLM: chat_template_kwargs
+# enable_thinking false): with thinking, some answers ran to 6-8k tokens,
+# minutes each, and stretched a test from 3.5 to 20 minutes.
+E2E_MODEL = "qwen3.8-27b-nothink"
 
 
 def e2e_roles() -> RoleMap:
@@ -59,16 +61,14 @@ class FreshRepo:
     push_key: Path  # this repository's deploy key, private half
 
 
-@contextlib.contextmanager
-def fresh_repo(
-    operator: GiteaClient, run: str, keys: Path, *, agent: str, seed: bool = True
-) -> Iterator[FreshRepo]:
-    """A repository of its own for one test run, as the operator would make a
-    project's (D4, D8): seeded with a small Python package, the ``agent``
-    account a collaborator with write access, the default branch protected (pushes only by the
-    owner, an approval to merge), and a deploy key of its own, written under
-    ``keys``. Deleted afterwards, unless NISSE_E2E_KEEP_REPOS is set."""
-    name = f"nisse-e2e-{run}"
+def make_repo(
+    operator: GiteaClient, name: str, keys: Path, *, agent: str, seed: bool = True
+) -> FreshRepo:
+    """A repository as the operator would make a project's (D4, D8): seeded
+    with a small Python package (or only a README), the ``agent`` account a
+    collaborator with write access, the default branch protected (pushes only
+    by the owner, an approval to merge), and a deploy key of its own, written
+    under ``keys``. A failure part-way deletes what was made."""
     full = f"{OWNER}/{name}"
     operator._request(
         "POST",
@@ -145,8 +145,35 @@ def fresh_repo(
                 "required_approvals": 1,
             },
         )
-        yield FreshRepo(full, key)
+        # The labels the agent reads (D1), as a project would have them.
+        for label, color in (
+            ("bug", "#ee0701"),
+            ("discuss", "#0e8a16"),
+            ("design", "#5319e7"),
+        ):
+            operator._request(
+                "POST", f"/repos/{full}/labels", json={"name": label, "color": color}
+            )
+    except BaseException:
+        delete_repo(operator, full)
+        raise
+    return FreshRepo(full, key)
+
+
+def delete_repo(operator: GiteaClient, full: str) -> None:
+    with contextlib.suppress(GiteaError):
+        operator._request("DELETE", f"/repos/{full}")
+
+
+@contextlib.contextmanager
+def fresh_repo(
+    operator: GiteaClient, run: str, keys: Path, *, agent: str, seed: bool = True
+) -> Iterator[FreshRepo]:
+    """A repository of its own for one test run (make_repo), deleted
+    afterwards unless NISSE_E2E_KEEP_REPOS is set."""
+    repo = make_repo(operator, f"nisse-e2e-{run}", keys, agent=agent, seed=seed)
+    try:
+        yield repo
     finally:
         if not os.environ.get("NISSE_E2E_KEEP_REPOS"):
-            with contextlib.suppress(GiteaError):
-                operator._request("DELETE", f"/repos/{full}")
+            delete_repo(operator, repo.name)

tests/integration/test_end_to_end_loop.py

@@ -239,6 +239,7 @@ class Operator:
         return self.transport.outbound[-1].text if self.transport.outbound else ""
 
 
+@pytest.mark.smoke
 def test_the_slice_end_to_end(
     running: tuple[RunningLoop, InMemoryTransport, str],
 ) -> None:
@@ -501,6 +502,7 @@ def test_a_two_milestone_feature(
         operator.tidy(issues, prs)
 
 
+@pytest.mark.smoke
 def test_a_conversation_becomes_filed_work(
     running: tuple[RunningLoop, InMemoryTransport, str],
 ) -> None:
@@ -636,6 +638,7 @@ def test_a_conversation_becomes_a_new_project(
         assert any(f.startswith("docs/plans/") for f in files), files
 
 
+@pytest.mark.smoke
 def test_a_question_leaves_the_restatement_up(
     running: tuple[RunningLoop, InMemoryTransport, str],
 ) -> None:

uv.lock

@@ -12,6 +12,7 @@ members = [
 [manifest.dependency-groups]
 dev = [
     { name = "pytest", specifier = ">=8" },
+    { name = "pytest-xdist", specifier = ">=3" },
     { name = "ruff", specifier = ">=0.6" },
 ]
 
@@ -46,6 +47,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/d1/d6/3965ed04c63042e047cb6a3e6ed1a63a35087b6a609aa3a15ed8ac56c221/colorama-0.4.6-py2.py3-none-any.whl", hash = "sha256:4f1d9991f5acc0ca119f9d443620b77f9d6b33703e51011c16baf57afb285fc6", size = 25335, upload-time = "2022-10-25T02:36:20.889Z" },
 ]
 
+[[package]]
+name = "execnet"
+version = "2.1.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/bf/89/780e11f9588d9e7128a3f87788354c7946a9cbb1401ad38a48c4db9a4f07/execnet-2.1.2.tar.gz", hash = "sha256:63d83bfdd9a23e35b9c6a3261412324f964c2ec8dcd8d3c6916ee9373e0befcd", size = 166622, upload-time = "2025-11-12T09:56:37.75Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ab/84/02fc1827e8cdded4aa65baef11296a9bbe595c474f0d6d758af082d849fd/execnet-2.1.2-py3-none-any.whl", hash = "sha256:67fba928dd5a544b783f6056f449e5e3931a5c378b128bc18501f7ea79e296ec", size = 40708, upload-time = "2025-11-12T09:56:36.333Z" },
+]
+
 [[package]]
 name = "h11"
 version = "0.16.0"
@@ -177,6 +187,19 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/24/25/1de2678b631f5a49215c6c96fff41ba892b0a34df68d6d80292b1b48aa7f/pytest-9.1.1-py3-none-any.whl", hash = "sha256:37a86b45efb9a47a61a36449063e8e18d0cab3161329fc099eb21783169c4f0c", size = 386536, upload-time = "2026-06-19T10:58:31.347Z" },
 ]
 
+[[package]]
+name = "pytest-xdist"
+version = "3.8.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "execnet" },
+    { name = "pytest" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/78/b4/439b179d1ff526791eb921115fca8e44e596a13efeda518b9d845a619450/pytest_xdist-3.8.0.tar.gz", hash = "sha256:7e578125ec9bc6050861aa93f2d59f1d8d085595d6551c2c90b6f4fad8d3a9f1", size = 88069, upload-time = "2025-07-01T13:30:59.346Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ca/31/d4e37e9e550c2b92a9cbc2e4d0b7420a27224968580b5a447f420847c975/pytest_xdist-3.8.0-py3-none-any.whl", hash = "sha256:202ca578cfeb7370784a8c33d6d05bc6e13b4f25b5053c30a152269fd10f0b88", size = 46396, upload-time = "2025-07-01T13:30:56.632Z" },
+]
+
 [[package]]
 name = "pyyaml"
 version = "6.0.3"