docs(agents): rename review loop to superintend, add convergence - #209
Open
haydenshively wants to merge 4 commits into
Open
docs(agents): rename review loop to superintend, add convergence#209haydenshively wants to merge 4 commits into
haydenshively wants to merge 4 commits into
Conversation
The skill covers more than review — the satisficing self-check, the clean-room agent, and the synthesis that turns findings into decisions — and "review loop" collided with `/review`, the `reviewer` agent, and `cross-check-codex`. - Self-Check becomes the first beat of Parameterization B, aimed at the provenance of the spec: a ticket handed over is not a reviewed intent, and a `create-issue` ticket's `Possible solution` is a suggestion to check against its own `Context`, not a spec to execute. - Parameterization A now says when Self-Check runs — against the request, as soon as the ask is fielded, before the draft exists rather than merely before the reviewers are dispatched. - Convergence is a new piece: report back to the reviewer and ask whether it accepts the resolution, since a cold re-read only re-derives round one. Stated once at its canonical symbol; A and B name it in their order lines, and C's thread resolution is labelled as the same piece in GitHub's medium. - Order lines use `+` for the steps that run in parallel. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
haydenshively
commented
Sep 10, 2026
Takes all four review suggestions on #209: broaden the description's first trigger to any task start, and trim the rationale from Convergence and from A's self-check timing. "Once you're confident in your draft" replaces "only once there is a draft" — the confidence bar is the real gate on dispatching. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
haydenshively
marked this pull request as ready for review
September 10, 2026 01:44
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7e01f5d8f7
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Synthesis's Decide bullet now says that incorporating means making the edit, so the piece ends with a changed artifact rather than a list of intentions. That covers A and B, which never named an apply step, without adding one to three order lines. Convergence's retry sentence sent a rejected artifact back to Independent Review before the objections were processed. The objections are findings: they re-enter Synthesis, and a fresh review is dispatched only when the artifact has moved far enough that nobody has read what is now there — the same bar C already applies to its conditional review. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A synthesis that drops or defers every finding is a legitimate outcome, so the piece cannot promise a changed artifact. The gloss on "incorporate" carries the point on its own. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
cashd
approved these changes
Sep 10, 2026
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
2 participants
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
Renames the
review-loopskill tosuperintendand closes two gaps in it.The skill had grown past its name. Its highest-value beat is the one that is not review — the satisficing self-check that asks whether the work needs to exist and whether the framing is right. "Review loop" also collided with three existing surfaces:
.claude/commands/review.md, therevieweragent, andcross-check-codex(which is literally the Codex-running part).Changes
Self-check before implementing. Parameterization B assumed a trustworthy source document — "You implement first, per the plan or TIB" — which a Linear ticket is not. Self-Check is now B's first beat, aimed at the provenance of the spec: say what you are implementing against, and treat a
create-issueticket'sPossible solutionas a suggestion to check against its ownContext, not a spec to execute. If the ticket is the only plan there is, it is an unreviewed draft, and running A on it may beat implementing it neatly. The frontmatterdescriptiongained the matching trigger, since none of the old ones fired on "here is a ticket, go build it".When the self-check runs, in A. The body said "after the self-check, dispatch these two agents", which read as though it happened just before dispatch. It runs against the request, as soon as the ask is fielded — before the draft exists.
Convergence. A new piece: tell the reviewer how you addressed its findings and ask whether it accepts them. A cold re-read of a changed artifact only re-derives round one, so the report-back is the work. Stated once at its canonical symbol, with the proportionality clause (one round with nothing further is convergence) and the exits (fresh session, re-run Self-Check, flag the difficulty). A and B name it in their order lines; C's thread-resolution rule is labelled as the same piece in GitHub's medium rather than duplicated.
Notation. Order lines use
+for steps that run in parallel:Not in scope
.claude/commands/execute-issue.mdis the actual ticket entry point and still references neither the skill nor the new self-check; its pre-push step invokesreviewer, which is Opus, and which the skill says is not an independent pass. Left alone deliberately — that flow is being reworked separately..claude/commands/review.md:275says "Terminate the review loop", meaning that command's own loop over findings. Untouched.Test plan
Docs and agent config only; no source files changed.
pnpm run format:check— cleanpnpm lint— 0 warnings, 0 errorspnpm knip— cleangrep -rn "review-loop\|review loop\|Review Loop"— only the unrelatedreview.md:275linemetadata.hermesnesting is intactpnpm testand typecheck skipped: zero TS/JS changed🤖 Generated with Claude Code