Blog Fact-Check
Verifies statistics, citations, and factual claims in blog posts.
The skill explicitly reads a blog and fetches cited URLs, with no stated credential access, external writes, or covert exfiltration, and it caps processing at 10 URLs per run. However, it does not address indirect prompt injection in fetched content, sensitive-data handling, user confirmation, permission isolation, or rollback, so 12 points are deducted.
The workflow, claim fields, scoring rules, sequential processing, and limitations are reasonably specified. However, extraction patterns are narrow, and diagnostics for nonstandard claims, redirects, network errors, and parsing failures are thin. Key paths were not executed in this static review, so the score is kept below the static-review ceiling.
Trigger phrases, file input, report output, and optional integration with blog-analyze are clearly described. However, the patterns are primarily English-oriented and the core workflow depends on WebFetch, with no Chinese-language coverage or fallback for mainland-China reachability and page-access constraints, so 6 points are deducted.
The file provides name, description, invocation metadata, argument hint, license, workflow, output example, limitations, and integration guidance. It lacks skill-specific versioning, changelog, maintenance ownership, update path, FAQ, and installation/dependency notes, so 7 points are deducted.
The task, extracted fields, confidence statuses, and recommended actions are concrete and could produce a directly reviewable report. However, fetch and matching behavior is not supported by skill-specific executable tests or committed sample results, and human review remains necessary, so the static cap of 7 applies and 8 points are deducted.
The scoring rubric, statuses, URLs, and locations provide some auditability. However, there is no third-party execution evidence, dedicated test suite, or independent corroboration for this skill; the supplied repository tests mainly cover other analysis and delivery components, so 6 points are deducted.
- Fetched page content may contain indirect prompt injection; treat it strictly as untrusted data and obtain human confirmation before processing potentially sensitive posts or URLs.
- No skill-specific execution tests are provided, so URL fetching, numeric matching, and abnormal-page handling are not verified in practice.
- Chinese statistical phrasing, Chinese source pages, and mainland-China reachability, paywalls, dynamic pages, and PDFs are limited or unverified.
- The claim to extract all statistical claims may exceed the documented pattern coverage; tables, charts, ranges, and implicit numeric claims may be missed.
What does this skill do, and when should you use it?
Blog Fact-Check is a focused verification skill within the claude-blog suite. It reads a blog file, identifies claims containing numbers, percentages, dollar amounts, or named sources, and checks cited URLs one by one. Using WebFetch, it searches source pages for the claimed data and assigns confidence based on exact, paraphrased, weak, or missing matches. Claims without source URLs are explicitly marked unverified.
Reads the target blog file and locates data claims; extracts claim text, numeric value, attribution, cited URL, and location; fetches cited pages sequentially with WebFetch; searches for the relevant value and surrounding context; assigns a confidence score and verification status; suggests search queries for uncited claims; and produces a report with claim-level results, summary counts, and recommended actions. It verifies no more than 10 URLs per run and marks the remainder as skipped.
- A content marketer checks percentages, monetary figures, and research claims before publishing an article.
- An editor confirms that statistics actually appear on the pages cited by a draft.
- A content team reviews a long article with many external references and needs an auditable verification table.
- A writer identifies unsupported factual claims and receives suggested searches for sourcing them.
What are this skill's strengths and limitations?
- Extracts and checks numerical claims, percentages, dollar figures, and named attributions individually.
- Uses explicit confidence scores from exact matches through not-found results.
- Flags claims without URLs instead of treating them as verified.
- Processes URLs sequentially and caps verification at 10 sources per run to reduce rate-limit pressure.
- It depends on accessible cited URLs; uncited claims cannot be directly verified.
- Paywalled, JavaScript-rendered, and PDF sources may require manual review or produce weak results.
- A 404 or unreachable URL receives a zero score, while finding a replacement source remains manual.
- The skill verifies claim-to-source correspondence; it is not a general writing, SEO, or editorial-audit tool.
How do you install this skill?
This skill is bundled in the public claude-blog repository. With Claude Code 1.0.33 or later, run:
/plugin marketplace add AgriciDaniel/claude-blog
/plugin install claude-blog@agricidaniel-claude-blog
Restart Claude Code after installation.
How do you use this skill?
Run this command in Claude Code:
/blog factcheck path/to/post.md
It can also be triggered with requests such as “fact check,” “verify statistics,” “check sources,” or “validate claims.” Provide the path to the blog file; the skill returns a verification report.
How does this skill compare with similar options?
Compared with asking a model to judge accuracy directly, this skill checks each cited page and records a match score. Compared with manual checking, it provides a structured claim list and verification report.