Plan 0012
One PR landing loop
One owned primitive that takes a PR from pushed to merged: state and checks polled together, failures fixed or surfaced, merge executed, superseded PRs disposed. Replaces the pr_watch, check-polling, and merge-script pile that forty-plus issues describe.
- Status
- Sketch
- Ambition
- 5 consolidation of existing watch, check, and merge machinery, not new theory
- Impact
- 8 the largest single issue cluster; every landed change pays the current tax
- Effort
- 6 one owned primitive replacing pr_watch, check polling, and merge scripts
- Risk
- 4 the old paths keep working while the loop absorbs them
- Maturity
- 1 sketch drafted from the issue corpus
- Leverage
- 8 every future PR, human or agent, rides the same loop
- Taste
- 4 failure modes are catalogued in forty issues; mostly disciplined engineering
- Authors
- Created
- Updated
- Tracking issue
- -
- Supersedes
- -
- Superseded by
- -
Summary
Landing a PR today involves pr_watch, gh pr checks --watch, manual state
polls, and merge commands that each miss states the others see: conflicting
PRs that skip CI, stale failed runs beside passing reruns, merges that land
without the watcher noticing, superseded drafts nobody closes. Over forty
open issues describe corners of the same machine. This plan builds that
machine once: a single landing loop that owns a PR from push to merged.
Non-goals
Changing merge policy (squash-only, required checks) or CI itself; the loop observes and acts, it does not redefine the gates.
Absorbed issues
#3494, #3168, #3387, #3349, #3306, #3280, #3281, #3147, #3114, #3097, #3373, #3369, #3359, #3319, #3138, #3002, #2973, #2948, #2855, #2849, #2846, #2839, #2818, #2816, #2815, #2885, #2897, #3054, #2279, #2254, #2075, #2198, #2149, #3029, #3028, #3027, #3026, #3025, #2914, #2913, #2831, #2860, #2857, #2834, #2812
Drafted by Claude Code (Fable) from the open-issue corpus; scores are seed estimates.
andrewgazelka