Fix lat.md errors in anticipation of 0.13 - #805
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
There was a problem hiding this comment.
ℹ️ Both fixes are correct and verified. One file with the identical defect was missed.
Reviewed changes — three docs-only edits under wiki/, no code or behavioral surface.
- Reference definitions split one-per-line —
wiki/research/cold-start.md:294andwiki/research/node-embedding-vs-spawn.md:130each packed every link reference definition onto one physical line; the PR breaks them apart. - Gitignored target delinked —
wiki/agents.md:23turns the markdown link to../AGENTS.local.mdinto a plain-text mention.
I confirmed the reflow is a genuine rendering fix rather than cosmetics. Parsing both forms through marked, the packed one-line form is not a valid link reference definition at all — CommonMark permits nothing but an optional title and whitespace after the destination before end-of-line — so the definition line renders as a literal text paragraph and every [label] reference in the body stays unlinked. The split form parses and resolves. The AGENTS.local.md delinking is likewise correct: the file is gitignored at .gitignore:140 and absent from a clean checkout, so that relative link always dangled.
Also verified on this branch: npm run lat:check passes, and no dangling relative markdown links remain anywhere in wiki/.
⚠️ A third file carries the identical packed-definition defect and is untouched
wiki/research/native-resolver-prior-art.md has the same one-line reference-definition blob at lines 48 and 141, so the sweep this PR performs is incomplete. It is not a cosmetic leftover: 27 labels are defined across those two lines and all 27 are referenced in the tables above them, meaning that page currently renders 27 broken references plus two walls of literal text. Since the goal is to be clean ahead of the checker upgrade, this file should land in the same pass.
Technical details
# Packed link reference definitions remain in `native-resolver-prior-art.md`
## Affected sites
- `wiki/research/native-resolver-prior-art.md:48` — 25 definitions packed onto one line
(`[PR 50322]` … `[marvinh.dev part-2]`). None parse; the line renders as a literal paragraph.
- `wiki/research/native-resolver-prior-art.md:141` — same defect with `[PR 54648]` and `[PR 55412]`.
## Required outcome
- Every reference definition in this file sits on its own line, matching the treatment
already applied to `cold-start.md` and `node-embedding-vs-spawn.md`.
- All 27 labels resolve to links in the rendered page.
## Verification
- Confirmed all 27 labels are referenced in body prose/tables, so each one is a live
broken reference rather than dead weight.
- Rendering behavior confirmed by parsing both forms with `marked`: packed definitions
emit `<p>[a]: <a …> [b]: <a …></p>` and leave `[a]` / `[b]` references unlinked.
## Suggested approach
- A repo-wide guard against regressions:
`grep -rnE '^\[[^]]+\]: *\S+ +\[[^]]+\]:' wiki/ --include=*.md`
returns exactly these two lines today and is empty once they are split.ℹ️ Nothing enforces these fixes yet — the checker pin is still 0.12.2
lat:check is pinned at package.json:44 to lat.md@0.12.2, which is the only version site in the repo (.github/workflows/lat-check.yml just shells out to npm run lat:check). So the errors this PR fixes are not currently caught by any gate, and the third file above will stay uncaught until the pin moves. Worth confirming whether the pin bump is intended as a follow-up in this PR or a separate one.
Technical details
# The `lat.md` version pin is unchanged
## Affected sites
- `package.json:44` — `"lat:check": "npx --yes lat.md@0.12.2 check"`, the sole version site.
- `.github/workflows/lat-check.yml` — invokes `npm run lat:check`; carries no pin of its own.
## Required outcome
- A decision on sequencing: bump the pin here so the fixes are enforced, or track the
bump separately and accept that `main` is unguarded against these errors until then.
## Open questions for the human
- `lat.md@0.13` is not yet on npm (`npm error code ETARGET` as of 2026-08-28), so I could
not empirically confirm which of these two error classes 0.13 actually flags. If the
intent is to land the doc fixes now and bump when 0.13 publishes, this is purely
informational.Claude Opus | 𝕏

No description provided.