Evidence — the ticket's reproduction, before and after

Work item: issue #172 · Testing plan row T11 (manual exploratory) · captured 2026-08-07, re-captured the same day after the owner-review rebuild to the pullRequests[] endpoint model.

The ticket's five reproduction steps, driven through a real Dispatcher with an injected FakeTmux (the same seam the integration tests use). The harness is reproduced verbatim at the end of this file rather than checked in as a script — a throwaway that exists to produce one capture belongs with that capture.

--prefix restores the pre-issue-172 behaviour by rebinding two methods — SessionRegistry.record_owning back to a bare find_by_work_item, and Dispatcher._record_pr_binding to a no-op — so both columns run the same code path everywhere else. No redaction was needed: the output holds github:octo/repo#N refs and file names.

Before — the defect

# BEFORE the fix (pre-issue-172 resolver)

## step 2 — a session is registered against the issue
registry: ['github-octo-repo-15.json']

## step 3 — a comment on PR #16, whose description says 'Closes #15'
routed refs: ['github:octo/repo#15', 'github:octo/repo#16']
delivered to: ['github:octo/repo#15']
spawned: []
registry: ['github-octo-repo-15.json']
record.pullRequests: []

## steps 4+5 — the link is removed; another comment on PR #16
routed refs: ['github:octo/repo#16']
delivered to: ['github:octo/repo#15']
spawned: []

Two lines carry the whole ticket:

• After step 3, record.pullRequests is empty. That is the ticket's own observation — "Note that sessions/ contains no file for the PR" — one representation later: the routing decision was made and discarded.
• After step 5, delivered to has not grown. The second comment was routed to github:octo/repo#16, which nothing on disk claims, and dropped. The issue's session is still running and still owns the work. (Under spawnOnUnmatched: always the same event would instead have started a duplicate session against the PR.)

After — the PR is on the record, with its own session

# AFTER the fix

## step 2 — a session is registered against the issue
registry: ['github-octo-repo-15.json']

## step 3 — a comment on PR #16, whose description says 'Closes #15'
routed refs: ['github:octo/repo#15', 'github:octo/repo#16']
delivered to: []
spawned: ['github:octo/repo#16']
registry: ['github-octo-repo-15.json']
record.pullRequests: ['github:octo/repo#16']

## steps 4+5 — the link is removed; another comment on PR #16
routed refs: ['github:octo/repo#16']
delivered to: ['github:octo/repo#16']
spawned: ['github:octo/repo#16']

The same lines, inverted — and one new behaviour:

• Step 3 records PR #16 on the issue's own record (record.pullRequests) and, because sessionPerPr defaults on, spawns the PR's own session with the comment as its boot prompt (spawned: ['github:octo/repo#16']). The registry still holds exactly one file.
• Step 5's routed refs is identical to the before-run (['github:octo/repo#16']): derivation still fails, exactly as the ticket describes. What changed is that the failure is no longer the end of the line — the second comment is delivered into the PR's recorded session (delivered to: ['github:octo/repo#16']), with no new spawn and no record ever minted for the PR.

The record on disk

{
  "workItem": {"ref": "github:octo/repo#15", "provider": "github",
               "owner": "octo", "repo": "repo", "number": 15},
  "harness": "claude",
  "harnessSessionId": "sess-1",
  "status": "active",
  "tmuxTarget": "loop-github-octo-repo-15",
  "pullRequests": [
    {
      "workItem": {"ref": "github:octo/repo#16", "provider": "github",
                   "owner": "octo", "repo": "repo", "number": 16},
      "harness": "claude",
      "harnessSessionId": "<uuid assigned at the endpoint spawn>",
      "status": "active",
      "tmuxTarget": "loop-github-octo-repo-16",
      "recentDeliveries": ["d-1", "d-2"]
    }
  ]
}

(Trimmed to the load-bearing fields; timestamps and the run's real uuid omitted.)

The harness

"""The ticket's reproduction, driven through a real Dispatcher.

Usage: repro.py [--prefix]   (--prefix restores the pre-issue-172 resolver)
"""

import sys, tempfile, time
from pathlib import Path

sys.path.insert(0, "cli")
sys.path.insert(0, "cli/tests")

from conftest import FakeTmux, StubInteractiveAdapter
from the_loop.control import ControlConfig
from the_loop.sessions import Session, SessionRegistry, WorkItemRef
from the_loop.webhook.dispatcher import Dispatcher, RoutingConfig
from the_loop.webhook.router import RoutedEvent, extract_work_items

PREFIX = "--prefix" in sys.argv
if PREFIX:
    SessionRegistry.record_owning = lambda self, ref: self.find_by_work_item(ref)
    Dispatcher._record_pr_binding = lambda self, routed, target: None

ISSUE, PR = "github:octo/repo#15", "github:octo/repo#16"


def pr_comment(body, pr_body):
    return {
        "action": "created",
        "repository": {"full_name": "octo/repo"},
        "issue": {
            "number": 16,
            "body": pr_body,
            "pull_request": {"html_url": "https://github.com/octo/repo/pull/16"},
        },
        "comment": {"body": body, "user": {"login": "octocat"}},
        "sender": {"login": "octocat"},
    }


def routed(body, pr_body, delivery):
    payload = pr_comment(body, pr_body)
    return RoutedEvent(
        event="issue_comment",
        action="created",
        delivery_id=delivery,
        work_items=extract_work_items("issue_comment", payload),
        payload=payload,
    )


tmp = Path(tempfile.mkdtemp())
local = tmp / "local"
registry = SessionRegistry(local)
tmux = FakeTmux()
dispatcher = Dispatcher(
    registry=registry,
    adapters={"claude": StubInteractiveAdapter()},
    config=RoutingConfig(control=ControlConfig(require_start_command=False)),
    tmux_runner=tmux,
)

print(f"# {'BEFORE the fix (pre-issue-172 resolver)' if PREFIX else 'AFTER the fix'}")
print()
print("## step 2 — a session is registered against the issue")
item = WorkItemRef.parse(ISSUE)
registry.register(
    Session(
        work_item=item,
        harness="claude",
        harness_session_id="sess-1",
        cwd=str(tmp),
        tmux_target=f"loop-{item.slug}",
    )
)
print("registry:", sorted(p.name for p in local.iterdir()))
print()

print("## step 3 — a comment on PR #16, whose description says 'Closes #15'")
event = routed("please rerun CI", "Closes #15", "d-1")
print("routed refs:", [r.ref for r in event.work_items])
dispatcher.handle(event)
time.sleep(0.5)
print("delivered to:", [ref for ref, _ in tmux.delivers])
print("spawned:", [ref for ref, *_ in tmux.spawns])
print("registry:", sorted(p.name for p in local.iterdir()))
rec = registry.find_by_work_item(ISSUE)
if rec is not None:
    print("record.pullRequests:", [pr.work_item.ref for pr in rec.pull_requests])
print()

print("## steps 4+5 — the link is removed; another comment on PR #16")
event = routed("and again please", "", "d-2")
print("routed refs:", [r.ref for r in event.work_items])
dispatcher.handle(event)
time.sleep(0.5)
print("delivered to:", [ref for ref, _ in tmux.delivers])
print("spawned:", [ref for ref, *_ in tmux.spawns])
dispatcher.stop()

Run as uv run python repro.py [--prefix] from the repository root. Two log lines about a missing gh binary and a workspace that is not a checkout of octo/repo are filtered from the captures above; both are the fixture's environment, not the behaviour under test.