docs: record why the reviewer registry could not be copied

[mlorentedev]

Lesson 096, from the #403 / #407 work. It merged before this was written; this carries the part worth keeping.

The finding

dotf pr triage-queue calls gh pr list --json comments. GitHub exposes a PR's discussion through three separate surfaces — issue comments, the reviews API (reviews[]), and inline review comments — and that call returns only the first.

On this repo CodeRabbit files proper reviews. #401's five findings live in reviews[] as **Actionable comments posted: 5**; comments held only the walkthrough. So registering the review's own text as the marker would have produced a queue that was permanently, silently empty on a repository that was being reviewed the whole time.

The dotfiles registry does not hit this because its second reviewer, PR-Agent, publishes through the comments API — which is why its marker does not transfer, and why "port the config" was the wrong mental model for #403.

Why it is worth a lesson rather than a commit message

The direction of failure. A wrong marker does not error — it returns "no reviewer output yet", which is also what a genuinely quiet PR looks like. Same family as 087 and 094: a check that cannot see its subject reports the subject is fine.

Two rules fall out:

1. When porting a config between repos, verify the observable it keys on against live data in the destination. One gh api repos/…/pulls/N/reviews versus …/issues/N/comments settles it in ten seconds; reading the source registry for an hour does not.
2. Prove a matcher in both directions. This one was confirmed by watching the queue correctly omit a triaged PR and correctly surface an untriaged one it found on its own — chore(master): release 4.2.0 #405, a real open release PR nobody had looked at, invisible until the registry existed.

Docs only. All wikilinks verified against the working tree.

[mlorentedev]

@coderabbitai review

[coderabbitai]

Warning

Review limit reached

Next included review available in 4 minutes.

View limit details

Limit details: You’ve used the included review currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 5d315e27-6e25-40e1-856f-a762c3105628

📥 Commits

Reviewing files that changed from the base of the PR and between 5fd5aa6 and 46a21b4.

📒 Files selected for processing (2)

• docs/lessons/_index.md
• docs/lessons/lesson-096-the-reviewer-publishes-where-the-consumer-does-not-read.md

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

⚠️ Action not completed

Review rate limited.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[mlorentedev]

Review triage

CI: all 6 check runs green (check 3.12/3.13, cross_worker_lock ubuntu + windows, docker, GitGuardian).

Review: none ran. "Review limit reached", commit status success. This PR is unreviewed.

Findings: none to dispose of.

Explicit review requested and it did not help. @coderabbitai review was issued on this PR, #405 and #408 at 03:27Z, right after the stated reset window. All three were acknowledged and none produced a review — reviews[] is still 0 on all three seven minutes later. That is the account-wide quota doing what dotfiles measured on 2026-08-16: roughly one review per hour across every repository under the account, so three PRs cannot be served no matter how they are staged.

Filed as #410 — hive runs one autonomous reviewer where dotfiles runs two, and pr-agent.yml exists there for precisely this. #403/#407 made the gap visible; it does not create capacity. Those are two problems and the second is now the binding one.

Content: docs only, one lesson file plus its index row. Every wikilink verified to resolve against the working tree.

Recommendation: merge on the human's call with the unreviewed status on the record, or hold until #410 lands if you would rather the lesson about review machinery be the first thing a second reviewer reads.

[coderabbitai]

Your plan includes PR reviews subject to rate limits. Reviews are available now.