wewrite-review — Editing, Revising and Finalizing Drafts
Prerequisites
When the user specifies an article, check that file; otherwise, run
, prioritize reading
, and also read
,
and
. Fall back to
only when there is no draft artifact for old tasks. First run
wewrite run step review in_progress
in the pipeline.
Full read list:
text
读取: {skill_dir}/../wewrite-write/references/article-brief.md
读取: {skill_dir}/../wewrite-write/references/editorial-quality.md
读取: {skill_dir}/references/seo-rules.md
When the user only says "check it" and provides an external file, only provide a report; when it is part of the main workflow, an in-task draft, or the user explicitly says "optimize", necessary revisions must be completed directly.
Review and Revision
1. Task Alignment
Check whether the article actually answers
, whether the core judgment is consistent with
, whether each section advances the specified
claim, and whether the conclusion delivers
. Delete irrelevant paragraphs, and do not use polished structure to cover up irrelevant answers.
2. Fact and Personal Material Verification
Run
wewrite sources list --json
, and cross-check specific numbers, dates, quotes, research conclusions, and time-sensitive facts in the text against
and original sources one by one. When sources do not support the content, prioritize supplementary verification; if it cannot be verified, delete it, narrow its scope, or rewrite it as an explicit inference. Do not use model memory to fill in gaps.
Check all first-person events, friends/colleagues, interviews, conversations, time/location, and sensory details. They must come from user materials explicitly recorded for this task; otherwise, they are blocking issues and should be directly deleted or rewritten as non-firsthand accounts.
3. Five-Dimensional Editorial Assessment
Rate each of the five dimensions — accuracy, viewpoint, usefulness, voice, and readability — on a scale of 1-5, and list blocking issues and up to 5 major issues:
- : Average score of at least 4, no single dimension below 3, and no blocking issues.
- : Can be corrected within existing materials. Make direct revisions and re-execute steps 1-3; do not only write suggestions.
- : Only applicable when the user explicitly requests personal stories but materials are insufficient and the framework cannot be safely adjusted.
Maximum two rounds. If there are still issues in the second round, delete unreliable content, reduce commitments, and generate a reliable version that can pass; do not label unapproved articles as "deliverable". After passing, write the final text to
.
4. Title, Abstract and Tool Tips
Generate one main title, two accurate alternative titles, an abstract within 40 characters, and 3-5 tags. Keywords should appear naturally, do not stuff them in based on density, and do not fabricate numbers or promises for click-through rates. Illustration allocation is not triggered during review.
Run
wewrite score {article_path} --json
. This score only helps identify risks such as clichés, overly uniform sentence structures, and paragraph rhythm issues; there is no mechanical passing line, and do not rewrite repeatedly just to increase the score.
Save Editorial Report
First write the editorial assessment as a temporary
file in the task directory:
yaml
decision: pass
pass_number: 1
dimensions:
accuracy: 4
viewpoint: 4
usefulness: 4
voice: 4
readability: 4
blockers: []
major_issues: []
notes: ""
bash
wewrite content-eval --draft {draft_path} --final {article_path} \
--assessment {assessment_path} --output {review_report_path} --json
The review can only be completed when
in the report. Update the task and mark it as completed:
bash
wewrite run update --patch '{"editorial":{"decision":"pass","pass_number":1,"publishable":true},"seo":{"title":"...","alt_titles":[],"digest":"...","tags":[],"quality_score":0},"provenance":{"verified_sources":0,"unverified_sources":0}}'
wewrite run step review completed
The self-check report should be in natural language, list up to 5 issues sorted by impact, and explain how they have been corrected. If there are no critical flaws, directly state that the text is deliverable; if needed, separate illustration or typesetting can be arranged later, and do not output lengthy score tables.