← Back to list
01Issue
FeatureShippedSwamp CLIPublic
Assigneesskunk-ape
Relationships
#4916 stagecraft: publish to the bound tracker after each write, and infer workItem
Opened by skunk-ape · 11/19/2026· Shipped 11/19/2026
Problem
A work item's ticket stays current only if the driver runs the publish workflow after every write. The skill says to (references/driving.md, "Keep the ticket in step"), and since 2771 status says when the tracker is behind. In practice the step costs a lot and still gets missed.
In factory sessions from Nov 4–18 (the cue and swamp-extensions factories):
- There were 67 manual
swamp workflow run @swamp/stagecraft/publishruns across 3 sessions. Each one is a separate command, and core's workflow renderer prints about 8K characters for it. - The first publish failed in each of those sessions because
--input workItem=<key>was left out. - A missed publish before a human gate leaves the ticket showing the wrong stage for as long as the person takes to answer, and the ticket is what reviewers read.
What to change
The driver should not have to run publish by hand after each write.
- Publishing follows writes without a separate driver step. After a write to a work item bound to a tracker, its events reach the ticket with no further command from the driver. A publish failure must not undo or fail the write itself: the write is recorded, the failure is reported, and the existing "tracker behind" line in
statusstill shows the lag. workItemis no longer a required input on publish. With noworkItem, publish delivers every work item on that tracker instance that has events waiting. That also gives a sweep or a driver tick something to call (DESIGN.md, "The publisher").
Constraints the plan has to respect
- The seam is deliberate: the engine knows nothing of trackers, and trackers are reached through the workflow contract (#3274, Route B). The plan should say how publishing follows a write without breaking that seam, or argue for changing it in a DESIGN decision entry. Options include a work-item write that runs the bound tracker's publish workflow, a wrapper workflow per write, or a sweep. Compare them and pick one.
- Publishing goes over the network and can be slow or fail. Say what happens to the write's latency, its output, and offline use.
- Built-in, swamp-club Lab and Linear trackers all have to keep working, and so does a work item started with no
externalRefs.
Done when
- Driving a work item through a bound factory needs no manual publish step, and the ticket is current at every human gate.
publishruns with noworkItemand delivers every work item with events waiting on that tracker.- A publish failure leaves the write recorded and says what failed.
statusstill shows the lag. - The skill and its driving reference drop the manual publish step, or say when it's still needed. README, REFERENCE and a DESIGN entry cover the change.
- Tests cover publishing after a write, publish without
workItem, a failed publish, and a work item with no tracker.
Out of scope
Short write output (#4907) and folding record_usage into other writes. This builds on #4907 and should start from main once that merges.
02Bog Flow
Shipped
Click a lifecycle step above to view its details.
03Sludge Pulse
Sign in to post a ripple.