Repository navigation
Add a parallel-review workflow and a PR #1 case study - #2
Merged
HenryCordes merged 2 commits intoJun 19, 2026
Merged
Conversation
The README claimed implementation "often" uses parallel subagents, but the repo had no actual example of that beyond one sequential test-writer agent - the claim had no evidence behind it. Adds .claude/workflows/parallel-review.js, a runnable multi-agent script that fans three reviewers (correctness, security, this project's own standards) out in parallel, then runs every finding through an independent skeptic before reporting it. The harness already discovers and parses it correctly (confirmed: it now shows up as an available workflow). Also adds docs/CASE_STUDY.md, a retrospective on PR #1 - the workflow's first real output on this repo - written to keep the actual friction (a fake-timer test race, a matcher that silently ignored a non-enumerable property) rather than presenting a sanitized success story.
Caught by actually running the workflow against this PR's own diff: the verify-stage label keyed only on finding.file, so two findings in the same file -- either multiple issues from one dimension, or the same file flagged by two of the three dimensions -- got identical labels while their verify agent() calls ran concurrently inside parallel(). pipeline() has no barrier between dimensions, so two dimensions' verify stages can genuinely overlap. Incorporates dimension.key and line/title so each verify call gets a label unique to the actual finding it's checking.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The README claimed implementation "often" uses parallel subagents, with nothing in the repo backing that up beyond one sequential test-writer agent. This adds a real artifact for that claim, plus an honest retrospective on the workflow's first PR.
Changes
.claude/workflows/parallel-review.js— a runnable multi-agent workflow: three reviewers (correctness, security, this project's own standards inAGENTS.md/docs/REACT_GUIDELINES.md) run in parallel againstgit diff <base>...HEAD, and every finding then goes through an independent skeptic agent before being reported. Verified the harness discovers and parses it correctly — it shows up as an available workflow after the file was added.docs/CASE_STUDY.md— a retrospective on PR Harden showcase tooling and add a second worked example #1, written to keep the real friction rather than a sanitized success story: a fake-timer test race (Promise.allracing the timer advance against the promise, instead of sequencing them) and atoMatchObjectthat silently ignored a non-enumerableError.causeproperty.eslint.config.js— scoped.claude/workflowsout of the app's lint config, same precedent as.husky(these are harness-context scripts, not application source).Testing
pnpm lint && pnpm typecheck && pnpm format:check && pnpm test— all green (21 tests, unchanged by this PR).parallel-review.js's syntax against the harness's actual execution model (theexport const metaliteral parses standalone; the rest of the body parses when wrapped in an async function, matching how the Workflow tool documents executing script bodies) — plainnode --checkisn't valid here since top-levelawait/returnare only legal inside that wrapper.Related
PROJ-000 (placeholder — no issue tracker on this personal showcase repo).
Checklist
pnpm lint:fix && pnpm typecheck && pnpm testpass locally{username}/{TICKET}-{description}