fix(gastown): production fixes, settings cleanup, and E2E testing by jrf0110 · Pull Request #2034 · Kilo-Org/cloud

jrf0110 · 2026-04-05T16:16:16Z

Summary

Production debugging, settings model cleanup, and E2E test infrastructure from the gastown-staging stabilization effort. Fixes several issues discovered during production monitoring of town 5f5fda7f.

Production fixes

Refinery dispatches despite code_review=false — When disabled, the reconciler now transitions ALL open MR beads to in_progress (not just those with pr_url), preventing the refinery from sneaking through the timing gap between MR bead creation and pr_url population.
Reconciler crash from NULL rig_id handling — Reverted the COALESCE sentinel approach that caused 500 errors on the dry-run endpoint. The MR beads actually had valid rig_id values (confirmed via debugGetBead).
Source bead closure guard — completeReviewWithResult only closes the source bead when it's in in_review state, preventing stale MR beads from incorrectly closing active source beads.
Direct-merge MR beads stuck — When code_review=false but MR beads have no pr_url (direct merge), the refinery is still dispatched for the merge operation.

Settings model cleanup

review_mode enum replaces code_review_comments boolean — 'rework' (internal gt_request_changes) or 'comments' (GitHub review comments on PR)
Convoy merge mode setting — Town-level default for new convoys: 'review-then-land' or 'review-and-merge'
Polecats always create PRs when merge_strategy: 'pr' (removed review-then-land convoy exception)
UI: Review mode button selector (disabled when merge strategy is direct), convoy merge mode selector, auto-merge delay input sized to match preset buttons

Code quality (from sub-agent review)

P0: poll_pr now calls checkPRFeedback once and reuses for both auto-resolve and auto-merge (halves GitHub API consumption)
P1: Rule 7 emits transition_agent to idle alongside stop+unhook (eliminates 1-tick inconsistent state)
P1: Removed duplicate metadata column from MR beads SQL query
P1: Clarified --approve prohibition in refinery prompt (GitHub limitation)
P2: PR comments review mode button disabled when merge strategy isn't PR

Debug infrastructure

GET /debug/towns/:townId/beads/:beadId — Full bead inspection (review_metadata, dependencies)
GET/PATCH /debug/towns/:townId/config — Read/update town config in dev
Debug bead endpoint accessible in production (behind CF Access, not dev-only gated)

Verification

pnpm typecheck — passes (full repo)
pnpm -r --filter cloudflare-gastown test — 181 tests pass
E2E Scenario 1: No review + auto-merge — PR feat: add Gateway section to sidebar and /gateway page #33 merged, auto-merge fired
E2E Scenario 2: Refinery review (rework) + auto-merge — PR skip balance checks in cloud-agent and webhook triggers #35 merged, full pipeline
E2E Scenario 3: Human feedback + auto-resolve + auto-merge — PR feat: use Opus for debug mode in auto model routing #37 merged, feedback detected and resolved
Production monitoring: town 5f5fda7f — reconciler 0 violations after deploy, refinery dispatching correctly

Visual Changes

Reviewer Notes

The review_mode enum replaces the short-lived code_review_comments boolean (which was only in the previous commit batch)
The convoy_merge_mode town config field uses convoy_ prefix to namespace it vs the per-convoy merge_mode column
The code_review=false gate now skips Rules 5-6 unconditionally — MR beads without pr_url are handled by orphan timeout rather than dispatching a refinery the user disabled

cloudflare-gastown/src/dos/Town.do.ts

cloudflare-gastown/src/dos/town/reconciler.ts

kilo-code-bot · 2026-04-05T16:28:55Z

Code Review Summary

Status: 1 Issue Found | Recommendation: Address before merge

Overview

Severity	Count
CRITICAL	0
WARNING	1
SUGGESTION	0

Fix these issues in Kilo Cloud

Issue Details (click to expand)

WARNING

File	Line	Issue
`services/gastown/src/gastown.worker.ts`	327	Malformed JSON still throws before Zod validation, so the debug config PATCH route returns a generic 500 instead of the intended 400.

Other Observations (not in diff)

N/A

Files Reviewed (4 files)

services/gastown/docs/e2e-pr-feedback-testing.md
services/gastown/scripts/prepare-container.mjs
services/gastown/src/dos/town/actions.ts
services/gastown/src/gastown.worker.ts - 1 issue

_{Reviewed by gpt-5.4-20260305 · 545,342 tokens}

cloudflare-gastown/src/prompts/refinery-system.prompt.ts

cloudflare-gastown/docs/e2e-pr-feedback-testing.md

…CF Access pattern The debug/beads/:beadId endpoint should be accessible in production (behind CF Access), matching the pattern of debug/status and other debug endpoints. The dev-only guard was incorrectly blocking production debugging.

…code_review is off When refineryCodeReview is disabled, only skip Rules 5-6 if all open MR beads have a pr_url. MR beads without pr_url (direct merge strategy) still need the refinery to merge, even when code review is disabled.

…les, add convoy merge mode setting - refinery.code_review: whether the refinery reviews PRs at all (gates + diff) - refinery.code_review_comments: whether the refinery posts GitHub review comments (only relevant when code_review is on). When off, the refinery uses internal rework requests instead of GitHub comments. - convoy_merge_mode: town-level default for new convoys — 'review-then-land' (single landing PR) or 'review-and-merge' (individual PR per bead). - Settings UI: conditional comments toggle under code review, convoy merge mode selector with review-then-land / review-and-merge buttons.

…ays create PRs Settings are now: 1. Refinery reviews (bool) — toggle refinery on/off entirely 2. Merge strategy (direct/pr) — already existed 3. Review mode (rework/comments) — how refinery communicates findings 4. Auto-resolve PR feedback (bool) — auto-dispatch polecat for comments/CI 5. Auto-merge + delay — auto-merge when PR is green - Replaced code_review_comments bool with review_mode enum ('rework'|'comments') - Polecats always create PRs when merge_strategy is 'pr' (removed review-then-land convoy exception — even intermediate beads get PRs) - UI: review mode is a button selector, shown only when code review is on - Convoy merge mode selector already added in previous commit

… transition, cleanup P0: poll_pr now calls checkPRFeedback once and reuses the result for both auto-resolve and auto-merge, halving GitHub API consumption. P1: Rule 7 now emits transition_agent to idle alongside stop+unhook, eliminating the 1-tick inconsistent working+unhooked state. P1: Removed duplicate metadata column from MR beads SQL query. P1: Clarified the --approve prohibition in the refinery prompt to explain it's a GitHub limitation (bot can't approve own PRs). P2: PR comments review mode button is disabled when merge strategy is not 'pr', with an explanatory tooltip.

…s (26x60px)

…lse, add debug config endpoints Bug fix: When code_review=false, the reconciler now transitions ALL open MR beads to in_progress (not just those with pr_url). This prevents the refinery from being dispatched during the timing gap between MR bead creation and pr_url population. MR beads without pr_url will be caught by the orphan timeout. Also adds debug GET/PATCH /debug/towns/:townId/config endpoints (dev only) for reading/updating town config without auth during local testing. Updates e2e-pr-feedback-testing.md with 3 test scenarios, pre-flight checklist, debug config endpoint docs, and known bugs.

…des, convoy target branch, direct-merge gate Thread 1: Always pass existingPrUrl to refinery prompt regardless of review_mode. The review mode now controls HOW findings are communicated (comments vs gt_request_changes) via the reviewMode param, not WHETHER the refinery knows the PR exists. Thread 2: Polecat prompt now resolves the convoy feature branch for review-then-land intermediate beads, so PRs target the convoy branch instead of main. Thread 3: Only fast-track MR beads with pr_url when code_review=false. Direct-merge MR beads (no pr_url) still dispatch the refinery for the merge operation via Rules 5-6.

The rework flow for existing PRs was using gt_mail_send which only delivers mail — it doesn't create the blocking rework bead needed to dispatch a polecat for fixes. Now uses gt_request_changes which correctly creates the rework bead and hands the MR back.

…sues

…essed

…n-null assertion

kilo-code-bot · 2026-04-06T03:23:13Z

services/gastown/src/gastown.worker.ts

+app.patch('/debug/towns/:townId/config', async c => {
+  if (c.env.ENVIRONMENT !== 'development') return c.json({ error: 'dev only' }, 403);
+  const townId = c.req.param('townId');
+  const parsed = TownConfigUpdateSchema.safeParse(await c.req.json());


WARNING: Malformed JSON still bypasses the new validation

safeParse(await c.req.json()) only runs after the request body has already been parsed. If a caller sends syntactically invalid JSON, c.req.json() throws before Zod sees it, so this debug endpoint still returns the worker's generic 500 instead of the 400 this validation path is trying to provide. Reusing parseJsonBody() here, like the authenticated config handler does, would keep bad debug requests from surfacing as server errors.

repoRoot was one level up (../), but after the move from cloudflare-gastown/ to services/gastown/ it needs to be two levels up (../../) to reach the workspace root.

jrf0110 · 2026-04-06T13:53:53Z

apps/web/src/app/(app)/gastown/[townId]/settings/TownSettingsPageClient.tsx

  const [refineryGates, setRefineryGates] = useState<string[]>([]);
  const [autoMerge, setAutoMerge] = useState(true);
  const [refineryCodeReview, setRefineryCodeReview] = useState(true);
+  const [reviewMode, setReviewMode] = useState<'rework' | 'comments'>('rework');


One day we'll migrate to https://react-hook-form.com/ and not have a bajillion useState's

jrf0110 self-assigned this Apr 5, 2026

kilo-code-bot bot reviewed Apr 5, 2026

View reviewed changes

cloudflare-gastown/src/dos/Town.do.ts Outdated Show resolved Hide resolved

cloudflare-gastown/src/dos/Town.do.ts Outdated Show resolved Hide resolved

cloudflare-gastown/src/dos/town/reconciler.ts Show resolved Hide resolved

kilo-code-bot bot reviewed Apr 5, 2026

View reviewed changes

cloudflare-gastown/src/prompts/refinery-system.prompt.ts Outdated Show resolved Hide resolved

jrf0110 changed the base branch from gastown-staging to main April 5, 2026 19:32

jrf0110 changed the base branch from main to gastown-staging April 5, 2026 19:32

jrf0110 changed the base branch from gastown-staging to main April 5, 2026 19:33

jrf0110 force-pushed the fix/prod-mr-bead-rig-id-and-debug branch from 5e0407e to d4dcf65 Compare April 5, 2026 19:36

kilo-code-bot bot reviewed Apr 5, 2026

View reviewed changes

cloudflare-gastown/docs/e2e-pr-feedback-testing.md Outdated Show resolved Hide resolved

jrf0110 added 11 commits April 5, 2026 22:10

style(settings): match auto-merge delay input height to preset button…

a7292a2

…s (26x60px)

docs(gastown): update e2e test docs with round 2 results and known is…

e6532c9

…sues

docs(gastown): mark known issues as historical, clarify they are addr…

34d98d6

…essed

jrf0110 force-pushed the fix/prod-mr-bead-rig-id-and-debug branch from 1183058 to 34d98d6 Compare April 6, 2026 03:12

fix: lint and format — validate debug config body with Zod, remove no…

621b82e

…n-null assertion

kilo-code-bot bot reviewed Apr 6, 2026

View reviewed changes

fix(gastown): fix container prepare script for monorepo directory depth

85780dd

repoRoot was one level up (../), but after the move from cloudflare-gastown/ to services/gastown/ it needs to be two levels up (../../) to reach the workspace root.

jeanduplessis approved these changes Apr 6, 2026

View reviewed changes

jrf0110 commented Apr 6, 2026

View reviewed changes

jrf0110 merged commit b30965c into main Apr 6, 2026
19 checks passed

jrf0110 deleted the fix/prod-mr-bead-rig-id-and-debug branch April 6, 2026 13:58

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(gastown): production fixes, settings cleanup, and E2E testing#2034

fix(gastown): production fixes, settings cleanup, and E2E testing#2034
jrf0110 merged 13 commits intomainfrom
fix/prod-mr-bead-rig-id-and-debug

jrf0110 commented Apr 5, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

kilo-code-bot bot commented Apr 5, 2026 •

edited

Loading

WARNING

Uh oh!

Uh oh!

Uh oh!

kilo-code-bot bot Apr 6, 2026

Uh oh!

jrf0110 Apr 6, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

jrf0110 commented Apr 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Production fixes

Settings model cleanup

Code quality (from sub-agent review)

Debug infrastructure

Verification

Visual Changes

Reviewer Notes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

kilo-code-bot bot commented Apr 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Code Review Summary

Overview

WARNING

Uh oh!

Uh oh!

Uh oh!

kilo-code-bot bot Apr 6, 2026

Choose a reason for hiding this comment

Uh oh!

jrf0110 Apr 6, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

jrf0110 commented Apr 5, 2026 •

edited

Loading

kilo-code-bot bot commented Apr 5, 2026 •

edited

Loading