Developer worker could not complete the task through `work_finish`

[netandreus]

Summary

A Developer worker completed implementation work, pushed a branch, and opened a merge request successfully, but could not complete the task through work_finish.

work_finish returned:

DEVELOPER worker not active on <Project_name>

This is here in the code: https://github.com/laurentenhoor/devclaw/blob/main/lib/tools/worker/work-finish.ts#L246

At the same time, DevClaw still showed an active developer worker slot for the same issue in project state, while task listing/status showed no active or queued issues at all.

This looks like a state reconciliation issue between:

• worker slot registry
• issue workflow state
• session visibility / session tree

---

Environment

• OpenClaw version: 2026.3.11
• DevClaw version: 1.6.10
• Workspace version: 1.6.10
• Channel: Telegram forum topic
• Project: <Project_name>
• Project slug: <Project_slug>
• Channel binding: -xxxxxxxxxx:topic:49

---

What happened

A Developer worker handled issue #2 and reported that it had:

• created test.txt with content Test passed
• committed changes
• pushed branch feature/2-create-test-txt
• opened MR:
  https://git.local.internal/group/project/-/merge_requests/5

However, when the worker attempted to call work_finish, it failed with:

DEVELOPER worker not active on <Project_name>

Retrying work_finish from the orchestrator session also failed with the same error.

Running health(fix=true) did not resolve the issue.

---

Observed inconsistent state

1. Project worker state says the developer worker is active

project_status reported:

• developer.activeSlots: 1
• active medior developer slot on:
• issueId: "2"
• recent startTime

devclaw/projects.json also contained an active developer slot with:

• issueId: "2"
• sessionKey: agent:main:subagent:project-slug-developer-medior-vyky

2. Task state says there are no issues at all

At the same time:

• tasks_status returned:
• totalHold: 0
• totalActive: 0
• totalQueued: 0

And:

• task_list(stateType="all") returned:
• totalIssues: 0

So DevClaw reported an active worker slot for issue #2, but no issue existed in any workflow state.

3. Session visibility was inconsistent

openclaw status showed that a subagent/direct session existed, consistent with the worker being alive.

But from the orchestrator session:

• sessions_list did not expose that worker session
• sessions_send failed with:

Session send visibility is restricted to the current session tree (tools.sessions.visibility=tree).

This suggests the worker session existed at runtime, but was no longer reachable from the current session tree.

---

Expected behavior

One of the following should happen:

1. If the worker slot is active for issue #2, work_finish from that worker should succeed.
2. If the issue is no longer valid / no longer active in workflow state, DevClaw should:

• clear the stale worker slot automatically, or
• report a more precise reconciliation error.

3. health(fix=true) should detect and repair this split-brain condition.

---

Actual behavior

• work_finish fails with:
• DEVELOPER worker not active on <Project_name>
• health(fix=true) reports no issues
• worker slot remains marked active
• task lists remain empty
• orchestrator cannot reach the worker session due to session tree visibility

---

Likely root cause

This appears to be a split-brain / stale state reconciliation bug involving at least one of:

• project worker slot registry (projects.json)
• issue workflow state / labels in the tracker
• session identity or session tree binding for the worker
• work_finish worker ownership validation

[netandreus]

I'm trying to debug this issue.
DevClaw's work_finish tool is refusing to complete because it believes "no active developer worker" is associated with this project/session even though the debug dump shows a junior worker slot is active.

Key fields in the debug block

project - name: "H1-Registrar (Tenant)" - slug:"h1-registrar-(tenant)"
role - "developer"
toolSessionKey (from the current work_finish call) - agent:main:subagent:h1-registrar (tenant)-developer-junior-shoshanna
activeSlots.junior[0] (the actually active worker slot)

{
  "index": 0,
  "active": true,
  "issueId": "6",
  "sessionKey": "agent:main:subagent:H1-Registrar (Tenant)-developer-junior-shoshanna",
  "startTime": "2026-03-13T12:11:14.867Z"
}

Why DevClaw says "worker not active"

Looking at work-finish.ts:

if (slots[i]!.active && slots[i]!.issueId &&
   (!toolCtx.sessionKey || !slots[i]!.sessionKey ||
    slots[i]!.sessionKey === toolCtx.sessionKey)) {
  // treat as the matching active slot
}

Important behavior

The plugin only accepts an active slot if the slot's sessionKey matches the tool's sessionKey (when both are present).

In my case

• toolSessionKey is all-lowercase at the start: agent:main:subagent:h1-registrar (tenant)-...
• The active slot's sessionKey starts with a capitalized project name:agent:main:subagent:H1-Registrar (Tenant)-...

Because the strings differ (case + slug/name mismatch), the condition slots[i]!.sessionKey === toolCtx.sessionKey evaluates to false, so the active slot is ignored.
After scanning all slots: slotIndex / issueId stay null so the tool throws: DEVELOPER worker not active on H1-Registrar (Tenant). even though activeSlots.junior[0].active == true

Root cause

A session key mismatch between:

• the worker slot
• the current work_finish call

Most likely caused by different normalization of the project name / slug:

• capitalization differences
• parentheses in " (Tenant)"

Behavioral meaning

You're calling work_finish from a context/session DevClaw does not consider the same worker that originally picked up the task.

This typically happens when

• work_finish is triggered from a different subagent/session
  than the one that started the work, or
• There is a bug in session key generation (case/slug mismatch for
  project names with spaces or parentheses).