Run the docs checks in CI and add a runnable-samples lint - #1380
Merged
Conversation
The Spanish, French and Portuguese trees translated the scrape format name changeTracking into seguimientoDeCambios, suiviDesModifications and rastreioDeMudancas. The API accepts only changeTracking, so every code sample carrying the translated name fails when a reader copies it, and the CLI --format tables list a flag value that does not exist. Inside code fences, inline code and quoted strings the name is now changeTracking again. Sentences that describe the feature rather than name the format keep prose in their own language. Finding OB-01 in DI-2026-09-04-WEEKLY reported this for French only. It is also present in Spanish and Portuguese. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SL5chNkWnr4Gy6uuB8PeKS
The Spanish webhooks page documents the event as crawl.started in its table and then sends "type": "rastreo.iniciado" in the JSON payload right below it. A handler written from that payload never matches the event the API sends. Finding OB-09 in DI-2026-09-05-WEEKLY, quoting the page: "type": "rastreo.iniciado" next to | `crawl.started` | El trabajo de rastreo comienza a procesarse |. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SL5chNkWnr4Gy6uuB8PeKS
The page declares const scrapedData in its first sample and const dadosRaspados in the second. The English page uses scrapedData in both. Translating an identifier inside a code sample makes the two snippets disagree for no benefit. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SL5chNkWnr4Gy6uuB8PeKS
Nothing in the repository stops a translation pass from rewriting an identifier inside a code sample, which is how changeTracking and crawl.started reached readers translated. scripts/check-locale-api-literals.sh compares the identifiers used inside code fences and inline code between each localized .mdx and its English counterpart. It reports an API identifier the English page uses in code and the localized page has dropped, and any identifier-shaped token a localized page uses in code that no English page uses anywhere. The vocabulary for the first check comes from the OpenAPI specs, so it follows the API instead of a hand-written list. Example domains are allowlisted. The repository has no CI workflows, so the script header says when to run it: before merging a change to the localized trees, and after every translation sync. It uses only git, jq and awk, the same tools as scripts/check-hosted-mcp-docs.sh, and takes about three seconds. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SL5chNkWnr4Gy6uuB8PeKS
The repository has no CI workflows, so the locale literal lint added in #1371 only runs when someone remembers to run it. This adds .github/workflows/checks.yml and a second script for it to run. The workflow uses pull_request with no branches filter. firecrawl-web's test.yml filters on branches: [main], which means a pull request stacked on another pull request's branch fires no event and gets no checks at all. The docs stack routinely does that, so the filter is left out. Permissions are contents: read, the trigger is pull_request rather than pull_request_target because the workflow runs code from the branch, and no secret is passed. Node 22 is pinned. scripts/check-extraction-hostile-markdown.sh checks that a published shell sample runs as written. An agent or a person who copies a sample out of the docs runs it as it stands, and a sample that uses a variable it never sets fails on the first line. The script flags a $VAR used inside a bash, sh or shell fence with no assignment in the same fence, allowing only $FIRECRAWL_API_KEY, which is the reader's own credential. Bare placeholders that are not shell variables, such as SCRAPE_ID in a URL path or fc-YOUR_API_KEY in a header, are not shell variables and are not flagged. It also flags raw i18n tokens and a backslash left inside a link target. It reads the English tree only, because the localized trees are regenerated and a finding there is fixed upstream. One rule was considered and dropped. A link anchor with a literal +, such as #persistent-profiles-with-scrape-+-interact, looks like an escaping artefact and is not one: the rendered heading on docs.firecrawl.dev carries exactly that id. Flagging it would fail a working link. Three findings are fixed here so the new check can gate: features/browser.mdx never set CDP_URL, and snippets/v2/monitor/check/get/curl.mdx never set MONITOR_ID or CHECK_ID. Findings: DI-2026-09-05-WEEKLY IP-D6 and IP-D8, reconciled plan items 12 and 14. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SL5chNkWnr4Gy6uuB8PeKS
hmishra2250
changed the base branch from
docs/keep-api-literals-untranslated
to
main
September 8, 2026 06:40
|
Preview deployment for your docs. Learn more about Mintlify Previews.
💡 Tip: Enable Automations to automatically generate PRs for you. |
# Conflicts: # fr/features/monitoring.mdx
…mat name The 2026-09-07 locadex pass rendered the `changeTracking` formats value as `suiviDesModifications`, which is not a value the API accepts. Restore the English literal so the sample runs as written.
hmishra2250
marked this pull request as ready for review
September 8, 2026 06:50
This was referenced Sep 8, 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
1 participant
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
Adds
.github/workflows/checks.yml, the repository's first CI workflow, and a second script for it to run alongside the locale literal lint from #1371.Opened as a draft. The new script currently fails on this branch, on purpose, and the reason is at the bottom under CI status.
Stack
Targets
docs/keep-api-literals-untranslated, the branch of #1371, because that PR shipsscripts/check-locale-api-literals.shand there is nothing for the workflow to run without it. Retarget tomainonce #1371 merges.Why
DI-2026-09-05-WEEKLY, paths IP-D6 and IP-D8, reconciled plan items 12 and 14.
There are no workflows in this repository at all.
git ls-tree -r origin/main --name-only .github/returns one PNG. The lint added in #1371 therefore runs only when someone remembers to run it, which is the same as not having it.The workflow deliberately does not copy the pattern in firecrawl-web's
.github/workflows/test.yml, which ispull_request: branches: [main]. A pull request stacked on another pull request's branch fires nopull_requestevent under that filter and gets no checks at all. This repository's docs stacks routinely target a sibling branch, this PR included, so the filter is left out and every pull request runs.Changes
.github/workflows/checks.yml. Triggerpull_requestwith nobranchesfilter.permissions: contents: read. Notpull_request_target, because the workflow runs code from the branch, and no secret is passed to it. Node 22 pinned withactions/setup-node@v4. Two steps, one per script. A concurrency group cancels superseded runs on the same pull request.scripts/check-extraction-hostile-markdown.sh. New. Checks that a published shell sample runs as written.scripts/check-locale-api-literals.sh. One comment line updated, from "This repository has no CI workflows" to a pointer at the workflow. No logic change.features/browser.mdxandsnippets/v2/monitor/check/get/curl.mdx. Three samples that used a variable they never set now set it.What the new script checks
$VARused inside abash,sh,shell,zsh,consoleorcurlfence with no assignment anywhere in the same fence.$FIRECRAWL_API_KEYis the only allowed name: it is the reader's own credential and every sample expects it in the environment. Assignments are recognised fromNAME=,export NAME=,for NAME inandreadwith any flags and any number of names. Positional parameters and the usual shell-provided names are not flagged.{locale}or a General Translation component such as<GTProvider>left in published markdown. Zero occurrences today, so this is a regression guard.The script reads the English tree only. The localized trees are regenerated by the translation pipeline, so a finding there is fixed in the pipeline or in the English source, never by hand editing a localized file, which
CLAUDE.mdforbids.Bare placeholders are not flagged, and the plan's caution about them does not apply.
SCRAPE_IDin a URL path andfc-YOUR_API_KEYin a header are literals with no$, so the rule never sees them. The two conventions do not collide anywhere in this corpus.One rule was checked and dropped. A link anchor containing a literal
+, such as#persistent-profiles-with-scrape-+-interact, looks like an escaping artefact and is not one.curl -s https://docs.firecrawl.dev/features/interact | grep -oE 'id="[^"]*persistent[^"]*"'returnsid="persistent-profiles-with-scrape-+-interact", and a+in a URL fragment is not decoded as a space. Flagging it would have failed a working link on #1364 and #1369, so it is not flagged and the script says why.Counts
scripts/check-locale-api-literals.sh:origin/mainat73103de7The 19 on main are what #1371 fixes. This branch inherits those fixes, so the lint is already green here.
scripts/check-extraction-hostile-markdown.sh:origin/mainat73103de7All 10 on main are undefined shell variables, and all 10 are real. Three are fixed here:
$CDP_URLinfeatures/browser.mdx:345, and$MONITOR_IDand$CHECK_IDinsnippets/v2/monitor/check/get/curl.mdx:2. The other seven are the$SCRAPE_IDuses in the interact curl snippets, which #1369 already fixes with its "Define SCRAPE_ID in the interact curl samples" commit. Neither branch can see the other's fixes, so the two sets are disjoint and complementary, verified by running this script inside #1369's worktree.CI status
Draft, because the new check fails on this branch with the seven
$SCRAPE_IDfindings that live on #1369's branch. It goes green once #1369 merges and this branch is rebased ontomain. Marking this ready before then would mean either a red required check or a check that does not gate.Making the check required
Turning this into a required status check is a repository settings change on the branch protection rule for
main, not something this PR can do. Do that after the first green run, using the check name "Locale literals and extraction-hostile markdown".🤖 Generated with Claude Code
https://claude.ai/code/session_01SL5chNkWnr4Gy6uuB8PeKS