Add review-doc for writing review of docs.page pages - #558
Open
BenInvertase wants to merge 20 commits into
Open
BenInvertase wants to merge 20 commits into
BenInvertase wants to merge 20 commits into
Conversation
|
🚅 Deployed to the docs.page-pr-558 environment in docs.page 2 services not affected by this PR
|
Do not require a job title or invent an audience. Title and description are enough, and recommended edits may only reuse words already on the page.
Skip first-mention expansion for interface names the same way PDF and HTTP already skip, so short how-tos are not forced to spell them out.
Six gold pages cover planted issues, skip-if idioms, fix-all, conflict grouping, mute, and review-each, with scripted assertions in evals.json.
One hit (and skip, when Skip-if exists) per writing scan. build_workers.py regenerates workers.json and fails if a check heading is missing.
Iteration runs, benchmarks, and the static viewer stay on disk and should not ship with the skill.
Keep the writing review distinct from the docs check CLI (links, assets, render). Publishers can say review-doc or review doc.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012j4ivvniwRPsyYqfYSqrfp
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012j4ivvniwRPsyYqfYSqrfp
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012j4ivvniwRPsyYqfYSqrfp
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012j4ivvniwRPsyYqfYSqrfp
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012j4ivvniwRPsyYqfYSqrfp
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012j4ivvniwRPsyYqfYSqrfp
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012j4ivvniwRPsyYqfYSqrfp
|
|
These two documentation fixes move to a separate PR at the author's request, so this branch stays focused on the review-doc skill. Reverts 5c838e2 and 0984081. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012j4ivvniwRPsyYqfYSqrfp
Member
There was a problem hiding this comment.
*Test
- Ran review-doc locally on the React Native Firebase getting-started page (~430 lines)
- Looked at the results table, then a full suggested rewrite side by side with the original
Learning
- I felt the checklist/table was good — easy to see which areas failed
- Long list of individual fixes below it was overwhelming.
- Side by side, a lot of the edits were tiny, so I had to look hard to spot what changed. Not wrong; the DX of using the output is the hard part
- I would think users would just hit “fix all” after a bit, because so many fixes are small (e.g.
e.g.→ “for example”), rather than accept/reject one by one.
Recommendations
- **Skill description ** add a short blurb on why choose this over other writing skills — e.g. you get a checklist and a results table first, then you decide what to change, instead of the agent quietly rewriting the page.
- Let me approve a kind of change (style vs structural) rather than only
tone-1,tone-2, … — that matches how I’d actually want to work when most edits are small
Group the violation list under each failing check and apply after the results print so the publisher can review a real diff. review each remains the opt-in pause.
Auto-applying a new heading id breaks inbound #slug links. Keep those findings for accept, and still apply sentence case when the id is unchanged.
Default apply still writes body edits and slug-stable sentence case; a slug-changing heading stays until accept. Review each leaves the same page untouched.
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
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 review-doc, a writing review for one docs.page page.
Technical writing is a profession. What makes a page easy to scan, act on, and trust is not common knowledge — mixed voice, vague links, padding, and unexplained jargon slip through even when the facts are right. Agents do not pick that up on their own. This skill is a utility that holds them to a best-practice bar: 16 writing checks, a results table after every run, then fix everything, decide one by one, or mute a check.
Say
use review-doc on docs/quickstart.mdx(or “review doc”). Usage:skills/review-doc/README.md.Eval results
We ran the same gold pages with the skill and without it. Fixtures:
skills/review-doc/evals/.Without the skill, agents often catch obvious wording issues. With it, they also follow the review (table, skip product idioms, group overlapping edits, mute, accept/reject).
<Info>Each of the 88 writing rules also has a tiny gold page (hit, and skip when the rule says skip). 172/172 passed.
Scope
app/(hosted site, MCP, Ask AI)packages/cli/packages/mdx-bundler/docs/(product documentation)Type of change
Test plan
bun run checkpasses locallyuse review-doc on docs/<page>.mdxprints results after every runfix all/review each/mute check <id>match the READMENotes for reviewers
Code is
skills/review-doc/. Live review logs are gitignored;logs/example.jsonis the schema.