Check upstream
Refreshes the Claude Code docs baseline, runs the conformance check, and recommends schema and rule updates. Use when asked to "check for upstream changes", "check schema drift", "are our schemas up to date", or "what is new in Claude Code".From its SKILL.md
npx -y skills add pdugan20/claudelint --skill check-upstreamAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 14 stars14 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
3.6 KB, as published. Nobody here has run it
Check Upstream
Detects new Claude Code documentation surface and drift between the official docs and claudelint's schemas, then recommends concrete updates.
The deterministic half of this system runs in CI. This skill adds the judgment the scripts deliberately do not attempt: reading the prose diff and deciding what it means for claudelint.
Usage
Run this skill when asked to check for upstream changes, verify schemas are up to date, or find out what changed in Claude Code's documentation.
Step 1: Refresh the baseline
Run npm run upstream:refresh.
Watch for two things in the output:
[NEW PAGE] <slug> (NOT WATCHED ...)- a docs page appeared that claudelint does not watch. This is how a whole feature (such as plugin dependencies) arrives. Read the page and decide whether it belongs insrc/upstream/watchlist.ts.Extractor guard tripped- the extractor stopped finding facts on a page. Upstream changed format. Fix the extractor in scripts/upstream/extract.ts. Do NOT lowerminFactsto make it pass; that disables the detector.
Step 2: Read the prose diff
Run git diff -- docs-baseline/.
This step is not optional, and it is not a skim. The scripts catch exactly two things: new pages, and hook-event drift. They do NOT catch field-level drift - a new, renamed, or removed field in any schema is caught ONLY by a human or agent reading this diff. They also cannot catch a constraint stated only in prose, and that class of change causes real bugs: the mintlify-docs install failure came from one sentence, "a bare string with only the plugin name".
So read the diff for both - changed fields and changed sentences. For each meaningful change, ask: does claudelint model this? Could a user violate it and have us report success?
Step 3: Run the conformance check
Run npm run check:upstream.
Its conformance findings cover hook events ONLY. A clean run means the HookEvents enum
matches the baseline - it says nothing about whether any other field is up to date. Field
drift is caught only by Step 2.
documented-not-modeled- upstream documents a hook event we do not model. Add it to the Zod schema, the manual schema inschemas/, and the website docs, then runnpm run check:schema-sync.modeled-not-documented- we model a hook event the docs do not document. Treat this as a probable hallucination. Verify against the official page before assuming it is real. If it is a deliberate claudelint extension, add it tosrc/upstream/extensions.tswith a reason. Never silence it by deleting the check.
Step 4: Report
Summarize: new pages, conformance findings, prose constraints worth a new rule, and a recommended action for each. Do not change schemas without saying what you are changing and why.
Important
- Never add a schema field that is not in the official docs. A previous session
hallucinated five skill frontmatter fields.
src/upstream/extensions.tsplus themodeled-not-documentedcheck exist to make that mistake mechanically visible. - Accepting a baseline refresh is a deliberate act: review
git diff -- docs-baseline/before committing it.
See also
docs/projects/upstream-watch.md- design and rationalesrc/upstream/watchlist.ts- watched pages and their minFacts guardsnpm run check:upstream- the offline conformance gate, also run on every PR