Repository navigation
Fleet reporting convention: queue task IDs and GitHub issue numbers collide — always prefix #1274
Description
Problem (encountered 2026-10-26)
Queue task IDs and GitHub issue numbers are both bare integers in the
same conversations, and they have started colliding: tonight queue
task 1233 (a code fix) and issue #1233 (epic(fleet)) were live
at the same time, and the operator's own delivery had to stop and
disambiguate. Shift reports, monitor summaries, and sweep.log lines
mix the two namespaces ("task 1175 merged" vs "#1131") with only
convention keeping them straight — and the convention slips (several
of tonight's shift reports needed parent-side reconstruction partly
because of it).
Fix (small)
Make the namespaces self-describing everywhere the fleet writes:
- Queue tasks are always rendered
task <id>(never a bare number,
never#<id>); GitHub issues/PRs are always#<n>. - Sweep.log, shift reports, monitor summaries, tracker, and the
repo-queue helper's own output follow the same rule. - The worker instruction and lifecycle protocol state the rule once,
with the collision as the worked example.
Acceptance criteria
- The rule is written into the worker instruction + lifecycle
protocol. - A spot-check of one day of sweep.log + shift reports shows no
bare queue-task numbers. - repo-queue.py output prefixes task ids consistently (or the
issue records why the helper is exempt).
Child of epic #1233. Filed by Muse (operator) 2026-10-26.
Metadata
Metadata
Assignees
Labels
Infrastructure, deployment, and operationsTrivial: expected under 30 minutes of workRoutine maintenance, housekeeping, dependency hygieneIntake review done: acceptance criteria, links, and normalized labels addedNice to have; do when convenientQueue task is parked (waiting on CI, time, or an event)Filed by the new-issue-creator jobDeliberately paused with a recorded resume conditionQueue packet kind: verify, reproduce, and route this itemQueue worker lifecycle state
Activity
developing-today commentedon Oct 27, 2026
CODE-398
50 remaining items
dezren39 commentedon Nov 3, 2026
🤖 Agent note · parked
dezren39 commentedon Nov 5, 2026
🤖 Agent note · picked up
dezren39 commentedon Nov 5, 2026
Spot-check — queue task 4650 (2026-11-04, lane code-h)
Read-only rolling 24-hour audit: 2026-11-04T00:27Z–2026-11-05T00:27Z. I scanned the rotated/active sweep-log segments covering the window (
sweep-archive-2026-11-04.log,sweep-2026-11-04-0755-archive.log,sweep-2026-11-04-1212-archive.log,sweep-2026-11-03-1942-archive.log, andsweep.log), deduped to 1,653 unique timestamped lines in the window.Using the same conservative task-list scan as the 2026-11-03 audit (plural
tasks <id>, <id>, …/tasks <id>/<id>constructions where every ID after the first is unprefixed), the window contains 81 records with 733 unprefixed queue-task IDs — worse than the previous window (task 4650's 2026-11-03 audit: at least 45 records / 289 IDs). Examples: 2026-11-04T00:44Z liststasks 7905, 8952, 7217, 1465, …(17 unprefixed IDs in one line); 2026-11-04T00:52Z lists 18; the pattern is pervasive across lane and monitor lines all day. These are task IDs in task lists, not counts, issue numbers, ports, or note IDs.Acceptance 1 is verified fresh: the worker instruction §9 and the lifecycle protocol §2 both carry the
task <id>/#<n>rule, the task 1233 vs issue #1233 worked example, and the field-labelled structured-JSON exemption for the queue helper. Acceptance 3 holds on this shift's evidence:fleet queueoutput is field-labelled JSON throughout. Acceptance 2 fails again.So #1274 stays open. The documented convention is not the gap — fleet writers (lane shift lines and monitor summaries in sweep.log) still emit plural task lists with bare IDs. Historical logs were not rewritten.
dezren39 commentedon Nov 5, 2026
🤖 Agent note · parked