reCode
issues in · reviewed PRs out · while you sleep
:~/app$ gh issue create -t "Rate-limit the public API" -l planned
#214 · labeled planned — that's the whole job description
:~/app$ logout
Connection closed. The loop is awake.
03:12:41 census open_prs=0 planned_issues=1 advance_ready=214
03:12:42 verdict action=advance issue=214 → driver spawned
03:13:20 scope 2 sub-tasks: middleware (api/) · config+docs — non-overlapping, one worker each
03:14:05 workers implementer×2 in isolated worktrees · branch feat/issue-214-rate-limit
03:41:17 review correctness lens: REJECT — "token bucket refills on wall clock, bursts drift under load"
03:52:03 fix implementer reworked refill to monotonic clock · re-review: approve
04:04:48 gates build ✓ lint ✓ types ✓ tests 214/214 ✓ coverage 91% ✓
04:06:31 pr #215 opened — bot-authored, so your approval is the merge key
05:22:09 ci-fix remote CI flake (matrix-only) → log tail fed back → fixed, re-pushed, green
07:58:00 idle cadence WATCH — waiting for the only human in the loop
:~/app$ gh pr review 215 --approve
✓ merged. The census already picked the next issue.
agents orchestrator scopes · implementers build in isolated worktrees · reviewers attack per lens
events a cron-less daemon: census → verdict → driver — a model runs only when there's work
reactions CI failures, review comments, sibling-merge conflicts — all fed back and fixed, bounded
ceilings attempt budgets + daily caps + needs-human escalation — it parks, it never spirals
sprints milestones gate the queue: the loop drains one version, you test, feedback becomes the next
:~/app$ /plugin marketplace add robercano/reCode && /plugin install orchestrator@recode
:~/app$ /orchestrator:setup
:~/app$ /pr-loop
:~/app$