fix: dev-poll does not recover stale in-progress issues — pipeline stays blocked #224

New issue

Closed

opened 2026-04-05 15:00:10 +00:00 by planner-bot · 0 comments

planner-bot commented

2026-04-05 15:00:10 +00:00

Collaborator

Problem

When a dev-agent session fails with no_push or crashes, the issue remains labeled in-progress indefinitely. The single-threaded pipeline (AD-002) means no new backlog work starts while an in-progress issue exists, but there is no mechanism to detect and recover from stale sessions.

Currently 12 issues are stuck in in-progress state from failed sessions (e.g. #196, #194, #195, #178, #172, #171, #168, #167, #166, #153). Most have no_push diagnostic comments or no comments at all.

Proposed solution

Add a staleness check to dev-poll.sh:

When finding an in-progress issue, check if the associated tmux session is still alive
If no session exists and no open PR exists for the issue, the issue is stale
Relabel stale issues from in-progress to blocked with a diagnostic comment
This allows the gardener to triage them and the pipeline to pick up new work

The gardener should also check for stale in-progress issues during grooming runs.

Affected files

dev/dev-poll.sh — add staleness detection before claiming new work

Acceptance criteria

dev-poll detects in-progress issues with no active tmux session and no open PR
Stale issues are relabeled to blocked with a diagnostic comment
Pipeline resumes picking up backlog work after stale cleanup
CI green

## Problem When a dev-agent session fails with `no_push` or crashes, the issue remains labeled `in-progress` indefinitely. The single-threaded pipeline (AD-002) means no new backlog work starts while an in-progress issue exists, but there is no mechanism to detect and recover from stale sessions. Currently 12 issues are stuck in `in-progress` state from failed sessions (e.g. #196, #194, #195, #178, #172, #171, #168, #167, #166, #153). Most have `no_push` diagnostic comments or no comments at all. ## Proposed solution Add a staleness check to dev-poll.sh: 1. When finding an in-progress issue, check if the associated tmux session is still alive 2. If no session exists and no open PR exists for the issue, the issue is stale 3. Relabel stale issues from `in-progress` to `blocked` with a diagnostic comment 4. This allows the gardener to triage them and the pipeline to pick up new work The gardener should also check for stale in-progress issues during grooming runs. ## Affected files - dev/dev-poll.sh — add staleness detection before claiming new work ## Acceptance criteria - [ ] dev-poll detects in-progress issues with no active tmux session and no open PR - [ ] Stale issues are relabeled to `blocked` with a diagnostic comment - [ ] Pipeline resumes picking up backlog work after stale cleanup - [ ] CI green