fix(workbench): give the workbench a ref so reviewer clones succeed by piekstra · Pull Request #565 · open-cli-collective/codereview-cli

piekstra · 2026-08-13T18:25:54Z

The symptom

On a real PR, four of five reviewers reported 0 findings and the rollup rendered it as a clean review. They had not found nothing — they had never run:

security:code-auditor       incomplete_failed
rust:implementation-tests   incomplete_failed
architecture:solid          incomplete_failed
policies:conventions        incomplete_failed
documentation:docs          complete_broad     ← the only agent needing no checkout

A reviewer that cannot start contributes zero findings, so a total failure of the reviewer fleet is indistinguishable from an all-clear. That is the part worth fixing regardless of the cause.

The cause

Prepare builds the workbench with init → fetch --no-tags <remote> <sha> → checkout --detach. Fetching by SHA writes only FETCH_HEAD, and a detached checkout writes no ref, so the resulting .git has no refs/:

workbench/.git:  FETCH_HEAD HEAD config description hooks index info logs objects
healthy clone:   HEAD config description hooks index info logs objects packed-refs refs

Git does not consider a directory without refs/ a repository, so the per-reviewer git clone --no-hardlinks <workbench> <workspace> fails with repository does not exist. Confirmed by hand: copying that workbench and writing a single ref makes the identical clone succeed.

The fix

git update-ref refs/heads/cr-review-head <head SHA> after the checkout. HEAD stays detached, so the pinned-checkout semantics are unchanged; the repo simply becomes clonable.

Test

TestPrepareLeavesWorkbenchClonable asserts the head is reachable through a real ref and then performs the actual git clone --no-hardlinks a reviewer workspace uses. Verified it fails without the fix and passes with it. Full suite green.

Prepare builds the workbench with init + fetch-by-SHA + checkout --detach, so the directory ends up with FETCH_HEAD and no refs/ at all. Git does not treat such a directory as a repository, so the per-reviewer git clone --no-hardlinks <workbench> <reviewer-workspace> fails with "repository does not exist" for every agent that needs a workspace. The failure is silent and actively misleading. A reviewer that cannot start is recorded as incomplete_failed and contributes **zero findings**, which the rollup then renders as a clean review. On a real PR this produced "0 findings" across four of five reviewers -- only the one agent that reads the diff without a checkout actually ran -- and the result read as an all-clear. Writing refs/heads/cr-review-head at the head SHA after checkout makes the workbench a valid, clonable repository. HEAD stays detached, so nothing else about the pinned checkout changes. The regression test asserts the property that matters -- that the workbench can actually be cloned the way a reviewer workspace is created -- and fails without the fix.

piekstra-dev

Automated PR Review

Reviewed commit: 992b1f3e4931
Profile: reviewer - Posting as: piekstra-dev

Summary

Reviewer	Findings
go:implementation-tests	0
harness-engineering:repo-health	0
security:code-auditor	0
architecture:solid	2

architecture:solid (2 findings)

Major - `internal/workbench/workbench.go:154`

U-L1 (behavioral contract): Prepare now establishes a new postcondition — the workbench is clonable because it has refs/heads/cr-review-head — but the reuse fast path does not check it. reusable (workbench.go:189-226) validates metadata, commitPresent for base/head, and verifyClean; none of those touch refs/, and metadataSchemaVersion is unchanged at 2 with schema 1 still accepted (workbench.go:204). A workbench built by a pre-fix binary therefore passes the gate verbatim, Prepare returns early, and every git clone --no-hardlinks in prepareReviewerWorkspace (workbench.go:416) still fails with "repository does not exist". That is precisely the silent zero-findings outcome this PR exists to fix, and it hits the most likely user: someone resuming the run whose reviewer fleet just died (NewestCompatibleIncompleteRun reuses the existing run's artifact dir, including workbench/).

Both paths out of Prepare should guarantee the same postcondition. Concrete fix: either add the ref check to the reuse gate — e.g. a deps.gitCommand(ctx, repoDir, "rev-parse", "--verify", workbenchHeadRef+"^{commit}") alongside the existing commitPresent calls, or fold it into verifyClean so build and reuse share one clonability assertion — or bump metadataSchemaVersion to 3 and stop accepting 1/2 in reusable, forcing a rebuild of workbenches that predate the ref. The check-based option is cheaper and keeps durable task fingerprints stable, which the schema-1 reuse note in docs/checkout-native-review-contract.md:106-109 explicitly cares about. Whichever you pick, the existing TestPrepareLeavesWorkbenchClonable should get a sibling that prepares a workbench, deletes refs/heads/cr-review-head, calls Prepare again, and asserts the result is clonable.

Minor - `internal/workbench/workbench.go:147`

U-G1 (new durable surface gets a documented contract): the workbench's ref set is now load-bearing shared state — workbench/repo/ is a durable run artifact whose shape docs/checkout-native-review-contract.md:44-80 and 90-109 specify, and the reviewer-clone consumer depends on it. The ref is described only in this inline comment. Worth an amendment to that doc's workbench-layout notes ('workbench/repo/ is a clean pinned checkout at the PR head SHA, carrying refs/heads/cr-review-head so the per-reviewer clone succeeds'), so the next change to Prepare does not re-lose it.

The same amendment should settle a second question this line decides implicitly: only the head gets a ref, and git clone transfers only ref-reachable objects, so the base commit that ensureCommit deliberately fetches (workbench.go:134) and that reusable asserts is present (workbench.go:215) does not survive into reviewer workspaces whenever base is not an ancestor of head — the normal case for a branch cut from an older main. Reviewers cannot then run git log base..HEAD or git diff <base> locally. This is defensible either way: reviewers are handed the provider-generated diff.patch and nothing in the pipeline reads base from the workspace, so it may simply be out of the workbench's job. If you want the base reachable, the fix is one more update-ref refs/heads/cr-review-base <base SHA> next to this call; if you do not, say so in the doc so the omission reads as a decision rather than an oversight. Evidence that would change this to a real defect: any reviewer prompt or workspace tool that resolves the base SHA inside workspaceRepo.

Reviewer Coverage

go:implementation-tests — complete (broad); skipped: none; constraints: none
harness-engineering:repo-health — complete (broad); skipped: none; constraints: none
security:code-auditor — complete (broad); inspected 1 assigned file (2 inspected across reviewers): internal/workbench/workbench.go; skipped: none; constraints: none
architecture:solid — complete (broad); inspected 1 assigned file (2 inspected across reviewers): internal/workbench/workbench.go; skipped: none; constraints: Only the assigned file internal/workbench/workbench.go was reviewed as a finding surface; workbench_test.go, runartifact, pipeline, and docs/checkout-native-review-contract.md were read for context only. Sandbox denied go test and git invocations in the reviewer workspace, so the fix and the reuse-path reasoning were verified by code reading rather than by running the suite. The PR's second stated symptom (a fleet-wide reviewer failure rendering as a clean review in the rollup) lives outside the assigned file and was not reviewed.

Inspected files (2)

internal/workbench/workbench.go
internal/workbench/workbench_test.go

0 PR discussion threads considered. 0 summarized; 0 resolved.

Completed in 5m 13s | $3.20 | claude-sonnet-5, claude-opus-5 | cr dev

Field	Value
Model	claude-sonnet-5, claude-opus-5
Reviewers	go:implementation-tests, harness-engineering:repo-health, security:code-auditor, architecture:solid
Engine	claude_cli · claude-sonnet-5, claude-opus-5
Reviewed by	cr · piekstra-dev
Duration	5m 13s wall · 7m 54s compute
Cost	$3.20
Tokens	116 in / 31.9k out

Per-workstream usage

orchestrator-selection — claude-sonnet-5
- In: 6
- Out: 2.6k
- Cache read: 66.7k
- Cache create: 17.7k
- Cost: $0.17
- Duration: 30s
go:implementation-tests — claude-sonnet-5
- In: 14
- Out: 4.5k
- Cache read: 246.6k
- Cache create: 46.2k
- Cost: $0.42
- Duration: 1m 00s
harness-engineering:repo-health — claude-sonnet-5
- In: 26
- Out: 5.7k
- Cache read: 568.1k
- Cache create: 48.4k
- Cost: $0.55
- Duration: 1m 15s
security:code-auditor — claude-sonnet-5
- In: 12
- Out: 2.2k
- Cache read: 183.7k
- Cache create: 24.5k
- Cost: $0.24
- Duration: 35s
architecture:solid — claude-opus-5
- In: 52
- Out: 16.0k
- Cache read: 1.1M
- Cache create: 64.9k
- Cost: $1.60
- Duration: 3m 48s
orchestrator-rollup — claude-sonnet-5
- In: 6
- Out: 965
- Cache read: 101.3k
- Cache create: 31.2k
- Cost: $0.23
- Duration: 43s

…e ref Review raised that Prepare's new postcondition -- the workbench is clonable because it carries a head ref -- was established only on the build path, while the reuse fast path returned early without checking it. Adding the assertion is right regardless, so it is here. But the specific scenario in the finding does not appear reachable: a workbench missing refs/ entirely is already rejected, because commitPresent and verifyClean shell out to git and fail in a directory git does not consider a repository. I tried several ways to construct a reuse that reaches a reviewer clone with a broken workbench and could not; deleting only the ref leaves refs/ in place, which git still accepts and clones fine. So the check stays as belt-and-braces for the narrower case (refs/ present, head ref gone) and to stop a future change to those checks quietly dropping clonability -- and the comment says exactly that rather than claiming to fix a failure I could not reproduce. No test accompanies it: a test that passes with the change reverted verifies nothing, and I would rather ship the guard honestly labelled than a green assertion that discriminates nothing. Also documents the ref in the checkout contract, per the same review. That includes the consequence the reviewer surfaced: only the head is given a ref, so the base commit does not reach reviewer workspaces when base is not an ancestor of head. Recorded as a decision with the one-line fix if it ever needs to change.

piekstra · 2026-08-13T19:15:16Z

Both findings addressed in d4b668f, one of them partly as a decline with reasons — flagging that explicitly so it can be overruled.

Major (reuse path): the assertion is added — both exits from Prepare should guarantee the same postcondition, and a future change to the existing checks shouldn't be able to drop clonability quietly.

But I could not reproduce the failure as described, and I want to be honest about that rather than let the commit imply otherwise. A workbench missing refs/ entirely is already rejected by the reuse gate: commitPresent and verifyClean shell out to git, and those fail in a directory git doesn't consider a repository. I tried it several ways — deleting only the head ref leaves refs/ in place, which git still accepts and clones from fine, so reusable returning true there is harmless.

So the check ships as belt-and-braces for the narrower case (refs/ present, head ref gone), and the comment says that rather than claiming a fix I couldn't demonstrate. No test accompanies it — I wrote one, confirmed it passed with the change reverted, and removed it. A test that doesn't discriminate reads as coverage while verifying nothing, which is the same class of problem as this PR's original bug.

If you can construct the reuse-reaches-clone case, I'll happily add the test and drop the hedging in the comment.

Minor (documentation): done. docs/checkout-native-review-contract.md now records the ref as load-bearing, and settles the second question you raised — only the head gets a ref, so the base commit doesn't reach reviewer workspaces when base isn't an ancestor of head. Written as a decision with the one-line change if it ever needs to be otherwise, rather than leaving it to read as an oversight.

Worth noting: this PR was reviewed by a cr binary built from this branch, because the released one hits the very bug being fixed. All four reviewers ran and none reported incomplete_failed — which is the fix demonstrating itself.

piekstra-dev

Automated PR Review

Reviewed commit: d4b668f4d386
Profile: reviewer - Posting as: piekstra-dev

Summary

Reviewer	Findings
go:implementation-tests	1
harness-engineering:repo-health	0
security:code-auditor	0
architecture:solid	0

go:implementation-tests (1 finding)

Major - `internal/workbench/workbench_test.go:724`

The reuse-path fix (workbench.go:230, refPresent check against workbenchHeadRef) has no test proving it, even though it guards the exact silent-zero-findings scenario this PR exists to fix. TestPrepareReusesValidV1WorkbenchWithoutRewritingMetadata only exercises reuse of a workbench that already has the ref (built by this same fixed code in the same test run), so it can't distinguish 'refPresent correctly gates reuse' from 'refPresent is a no-op'. Nothing in the suite deletes refs/heads/cr-review-head from an otherwise-valid, metadata-matching workbench and asserts Prepare rebuilds it into a clonable state rather than reusing it verbatim -- which is precisely the pre-fix-binary/resumed-run scenario the reuse check was added for. Add a sibling to TestPrepareLeavesWorkbenchClonable that prepares a workbench, runs update-ref -d refs/heads/cr-review-head (or equivalent) against the resulting repo to simulate a workbench built before this fix, calls Prepare again with an unchanged Request, and asserts (a) Prepare did not skip rebuilding via the reuse fast path and (b) the resulting workbench is clonable via the same git clone --no-hardlinks assertion already used in TestPrepareLeavesWorkbenchClonable. Without this, a future refactor of reusable() or refPresent() that silently breaks the gate would pass the full suite.

Reviewer Coverage

go:implementation-tests — complete (constrained); inspected 2 assigned files (3 inspected across reviewers): internal/workbench/workbench.go, internal/workbench/workbench_test.go; skipped: none; constraints: none
harness-engineering:repo-health — complete (constrained); skipped: none; constraints: none
security:code-auditor — complete (constrained); inspected 1 assigned file (3 inspected across reviewers): internal/workbench/workbench.go; skipped: none; constraints: none
architecture:solid — complete (constrained); inspected 1 assigned file (3 inspected across reviewers): internal/workbench/workbench.go; skipped: none; constraints: Both findings from the prior round are resolved at d4b668f: the reuse gate now asserts the head ref (workbench.go:230), and the doc amendment records the ref plus the deliberate head-only decision (checkout-native-review-contract.md:77-85). Findings are limited to the assigned file internal/workbench/workbench.go; workbench_test.go and docs/checkout-native-review-contract.md changed in this PR and were read for context only. Sandbox denied go test and git invocations from this workspace, so the reuse-path and clonability reasoning is from code reading, not from running the suite.

Inspected files (3)

docs/checkout-native-review-contract.md
internal/workbench/workbench.go
internal/workbench/workbench_test.go

0 PR discussion threads considered. 0 summarized; 0 resolved.

Completed in 2m 22s | $3.43 | claude-sonnet-5, claude-opus-5 | cr dev

Field	Value
Model	claude-sonnet-5, claude-opus-5
Reviewers	go:implementation-tests, harness-engineering:repo-health, security:code-auditor, architecture:solid
Engine	claude_cli · claude-sonnet-5, claude-opus-5
Reviewed by	cr · piekstra-dev
Duration	2m 22s wall · 5m 05s compute
Cost	$3.43
Tokens	80 in / 20.4k out

Per-workstream usage

go:implementation-tests — claude-sonnet-5
- In: 22
- Out: 6.3k
- Cache read: 849.3k
- Cache create: 75.2k
- Cost: $0.80
- Duration: 1m 23s
harness-engineering:repo-health — claude-sonnet-5
- In: 22
- Out: 5.9k
- Cache read: 825.2k
- Cache create: 82.8k
- Cost: $0.83
- Duration: 1m 23s
security:code-auditor — claude-sonnet-5
- In: 12
- Out: 1.4k
- Cache read: 224.8k
- Cache create: 39.1k
- Cost: $0.32
- Duration: 23s
architecture:solid — claude-opus-5
- In: 18
- Out: 5.2k
- Cache read: 644.4k
- Cache create: 74.2k
- Cost: $1.19
- Duration: 1m 27s
orchestrator-rollup — claude-sonnet-5
- In: 6
- Out: 1.7k
- Cache read: 112.9k
- Cache create: 37.3k
- Cost: $0.28
- Duration: 27s

piekstra-dev

Automated PR Review

Reviewed commit: 0f6dfb29f1d3
Profile: reviewer - Posting as: piekstra-dev

Summary

Reviewer	Findings
go:implementation-tests	0
harness-engineering:repo-health	0
security:code-auditor	0
architecture:solid	0

Reviewer Coverage

go:implementation-tests — complete (constrained); inspected 2 assigned files (3 inspected across reviewers): internal/workbench/workbench.go, internal/workbench/workbench_test.go; skipped: none; constraints: none
harness-engineering:repo-health — complete (constrained); skipped: none; constraints: none
security:code-auditor — complete (constrained); inspected 1 assigned file (3 inspected across reviewers): internal/workbench/workbench.go; skipped: none; constraints: none
architecture:solid — complete (constrained); inspected 1 assigned file (3 inspected across reviewers): internal/workbench/workbench.go; skipped: none; constraints: All three prior findings are resolved at 0f6dfb2: reuse gate asserts the head ref (workbench.go:230), the doc records the ref and the head-only decision, and TestPrepareRestoresHeadRefOnReuse covers the gate's rejection branch. Findings are scoped to the assigned file internal/workbench/workbench.go; workbench_test.go and docs/checkout-native-review-contract.md also changed and were read for context only. Sandbox denied go test and git execution from this workspace, so the diff was judged by code reading rather than by running the suite. The comment at workbench.go:225-229 claims a workbench with no refs/ is already rejected by commitPresent/verifyClean. Consistent with the PR's own findings and the author's reproduction attempt, but not independently verified here since git could not be run.

Inspected files (3)

docs/checkout-native-review-contract.md
internal/workbench/workbench.go
internal/workbench/workbench_test.go

0 PR discussion threads considered. 0 summarized; 0 resolved.

Completed in 1m 39s | $3.42 | claude-sonnet-5, claude-opus-5 | cr dev

Field	Value
Model	claude-sonnet-5, claude-opus-5
Reviewers	go:implementation-tests, harness-engineering:repo-health, security:code-auditor, architecture:solid
Engine	claude_cli · claude-sonnet-5, claude-opus-5
Reviewed by	cr · piekstra-dev
Duration	1m 39s wall · 2m 25s compute
Cost	$3.42
Tokens	52 in / 8.6k out

Per-workstream usage

go:implementation-tests — claude-sonnet-5
- In: 12
- Out: 2.0k
- Cache read: 532.4k
- Cache create: 87.5k
- Cost: $0.71
- Duration: 28s
harness-engineering:repo-health — claude-sonnet-5
- In: 16
- Out: 2.6k
- Cache read: 688.2k
- Cache create: 96.8k
- Cost: $0.83
- Duration: 38s
security:code-auditor — claude-sonnet-5
- In: 6
- Out: 743
- Cache read: 144.9k
- Cache create: 48.1k
- Cost: $0.34
- Duration: 12s
architecture:solid — claude-opus-5
- In: 12
- Out: 2.8k
- Cache read: 502.9k
- Cache create: 92.3k
- Cost: $1.24
- Duration: 54s
orchestrator-rollup — claude-sonnet-5
- In: 6
- Out: 537
- Cache read: 125.0k
- Cache create: 40.3k
- Cost: $0.29
- Duration: 11s

The previous commit added the reuse-path check without a test and said so, because the test I wrote passed with the change reverted. Review pointed out the reason, and it was a flaw in my assertion rather than in the scenario: I was asserting the reused workbench is clonable, which does not discriminate, since deleting only the ref leaves refs/ in place and git still clones from it happily. The discriminating assertion is that the ref comes back. Without the gate, Prepare takes the reuse path and the ref stays deleted; with it, the workbench is rebuilt and the ref resolves to the head SHA again. Confirmed red with the gate removed and green with it. Comment updated to describe what the check actually catches -- refs/ present but the head ref absent, which every other reuse check accepts -- rather than hedging about a scenario I had not managed to construct.

piekstra-dev

Automated PR Review

Reviewed commit: 4cfc86cd4b9a
Profile: reviewer - Posting as: piekstra-dev

Summary

Reviewer	Findings
go:implementation-tests	0
harness-engineering:repo-health	0
security:code-auditor	0
architecture:solid	0

Reviewer Coverage

go:implementation-tests — complete (constrained); inspected 2 assigned files (3 inspected across reviewers): internal/workbench/workbench.go, internal/workbench/workbench_test.go; skipped: none; constraints: none
harness-engineering:repo-health — complete (constrained); skipped: none; constraints: none
security:code-auditor — complete (constrained); inspected 1 assigned file (3 inspected across reviewers): internal/workbench/workbench.go; skipped: none; constraints: none
architecture:solid — complete (constrained); inspected 1 assigned file (3 inspected across reviewers): internal/workbench/workbench.go; skipped: none; constraints: All three prior findings remain resolved: build path sets the ref (workbench.go:154), the reuse gate asserts it (workbench.go:230), TestPrepareRestoresHeadRefOnReuse covers the rejection branch, and the contract doc records the head-only ref decision. Findings are scoped to the assigned file internal/workbench/workbench.go; workbench_test.go and docs/checkout-native-review-contract.md also changed and were read for context only. Sandbox denied go test and git execution from this workspace, so the diff was judged by code reading rather than by running the suite. The comment at workbench.go:225-229 claims a workbench with no refs/ is already rejected by commitPresent/verifyClean. Plausible and author-verified, but not independently confirmed here since git could not be run. internal/workbench/workbench.go is unchanged at 4cfc86c relative to the previously reviewed 0f6dfb2 (still +31 -2); only the test file grew, refactoring the clone assertion into cloneWorkbench/gitCommandSucceeds helpers.

Inspected files (3)

docs/checkout-native-review-contract.md
internal/workbench/workbench.go
internal/workbench/workbench_test.go

0 PR discussion threads considered. 0 summarized; 0 resolved.

Completed in 1m 23s | $3.83 | claude-sonnet-5, claude-opus-5 | cr dev

Field	Value
Model	claude-sonnet-5, claude-opus-5
Reviewers	go:implementation-tests, harness-engineering:repo-health, security:code-auditor, architecture:solid
Engine	claude_cli · claude-sonnet-5, claude-opus-5
Reviewed by	cr · piekstra-dev
Duration	1m 23s wall · 2m 00s compute
Cost	$3.83
Tokens	48 in / 7.1k out

Per-workstream usage

go:implementation-tests — claude-sonnet-5
- In: 12
- Out: 1.9k
- Cache read: 593.3k
- Cache create: 100.1k
- Cost: $0.81
- Duration: 29s
harness-engineering:repo-health — claude-sonnet-5
- In: 12
- Out: 2.1k
- Cache read: 645.7k
- Cache create: 109.9k
- Cost: $0.88
- Duration: 30s
security:code-auditor — claude-sonnet-5
- In: 6
- Out: 668
- Cache read: 162.6k
- Cache create: 56.8k
- Cost: $0.40
- Duration: 11s
architecture:solid — claude-opus-5
- In: 12
- Out: 2.0k
- Cache read: 593.4k
- Cache create: 108.2k
- Cost: $1.43
- Duration: 38s
orchestrator-rollup — claude-sonnet-5
- In: 6
- Out: 443
- Cache read: 129.1k
- Cache create: 43.3k
- Cost: $0.31
- Duration: 10s

…566) ## The problem Zero findings and "did not run" are the same number and opposite meanings. One says the code was examined and is clean; the other says nothing was examined. The summary table rendered both as `0`. On a real PR, this is what a run looked like where **four of five reviewers failed to start**: ``` | Reviewer | Findings | | security:code-auditor | 0 | | rust:implementation-tests | 0 | | architecture:solid | 0 | | policies:conventions | 0 | | documentation:docs | 1 | ``` Only `documentation:docs` actually ran — it is the one agent that needs no repo checkout. The failures *were* recorded, but as `incomplete_failed` in a coverage section well below the headline table. The summary is the part that gets read, pasted, and acted on, and it said all-clear. I nearly merged on that reading, twice. ## Scope **The outcome logic is already correct** — `hasIncompleteReviewerCoverage` coerces approval away when coverage is incomplete, so the PR was never actually approved. This change is purely about what a human reads. ## The fix A reviewer whose coverage status did not produce a result renders as `⚠️ did not run` instead of a count, reusing the existing `coverageResultProduced` helper. Reviewers **absent** from coverage keep their number, so an unknown status cannot turn a genuine zero into a phantom failure. That case has its own test. ## Tests Two tests, both confirmed to **fail with the change reverted**: - a failed reviewer is not rendered as `0`, while a genuinely clean reviewer keeps its honest `0` in the same table - unknown coverage leaves the count untouched Full suite green. ## Related [#565](#565) fixes the *cause* of that particular fleet failure (the workbench has no `refs/`, so per-reviewer clones fail). This one fixes the reporting, which matters for any reviewer failure regardless of cause — the two are independent.

piekstra mentioned this pull request Aug 13, 2026

fix(rollup): a reviewer that never ran must not report zero findings #566

Merged

piekstra-dev reviewed Aug 13, 2026

View reviewed changes

Comment thread internal/workbench/workbench.go

Comment thread internal/workbench/workbench.go

piekstra-dev reviewed Aug 13, 2026

View reviewed changes

Comment thread internal/workbench/workbench_test.go

piekstra-dev previously approved these changes Aug 13, 2026

View reviewed changes

piekstra dismissed piekstra-dev’s stale review via b490d78 August 13, 2026 19:26

piekstra force-pushed the fix/workbench-refs branch from b490d78 to 4cfc86c Compare August 13, 2026 19:30

piekstra-dev approved these changes Aug 13, 2026

View reviewed changes

piekstra merged commit a225189 into main Aug 13, 2026
10 checks passed

piekstra deleted the fix/workbench-refs branch August 13, 2026 19:35

Navigation Menu

Uh oh!

fix(workbench): give the workbench a ref so reviewer clones succeed - #565

fix(workbench): give the workbench a ref so reviewer clones succeed#565
piekstra merged 3 commits into
mainfrom
fix/workbench-refs

piekstra commented Aug 13, 2026

Uh oh!

piekstra-dev left a comment

Uh oh!

Uh oh!

Uh oh!

piekstra commented Aug 13, 2026

Uh oh!

piekstra-dev left a comment

Uh oh!

Uh oh!

piekstra-dev left a comment

Uh oh!

piekstra-dev left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

piekstra commented Aug 13, 2026

The symptom

The cause

The fix

Test

Uh oh!

piekstra-dev left a comment

Choose a reason for hiding this comment

Automated PR Review

Summary

Major - internal/workbench/workbench.go:154

Minor - internal/workbench/workbench.go:147

Reviewer Coverage

Uh oh!

Uh oh!

Uh oh!

piekstra commented Aug 13, 2026

Uh oh!

piekstra-dev left a comment

Choose a reason for hiding this comment

Automated PR Review

Summary

Major - internal/workbench/workbench_test.go:724

Reviewer Coverage

Uh oh!

Uh oh!

piekstra-dev left a comment

Choose a reason for hiding this comment

Automated PR Review

Summary

Reviewer Coverage

Uh oh!

piekstra-dev left a comment

Choose a reason for hiding this comment

Automated PR Review

Summary

Reviewer Coverage

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Major - `internal/workbench/workbench.go:154`

Minor - `internal/workbench/workbench.go:147`

Major - `internal/workbench/workbench_test.go:724`