Last checked October 24, 2026 · Component files read line by line on October 19, 2026 at commit 161d491, re-checked on October 23 at commit 7d85535 (no change) and on October 24 at commit 72e5f23 (only the seoagent skill's frontmatter name changed). aitmpl.com pages and their scanner badges read on October 23, 2026. Nothing from the catalog was installed or run.
No affiliate links on this page. Saazora earns nothing from Claude Code Templates, its author, NVIDIA or any vendor named below.
This Claude Code Templates review comes down to one number: of eight marketing components read in full, two are worth installing as they are, four are worth having only after you read their files, and two are better skipped. Claude Code Templates, the catalog at aitmpl.com, describes itself on its homepage as "an open-source catalog of 1,912+ components for Claude Code". Saazora did not try to cover all of it. We picked the eight components an SEO or affiliate marketer is most likely to reach for, opened every file in each, and wrote down the exact lines that decide the verdict.
If you are new to Claude Code, start from Saazora's AI Agents & Claude Code hub, then come back here. The short version for people in a hurry: install seo-audit and competitor-alternatives by copying their SKILL.md by hand, read four more before you let them near a project, and leave publisher-linkedin and ai-directory-submitter alone for now.
The Claude Code Templates review in one table: eight verdicts
The table is the review in one place. "Runs or reaches out" lists only what the files themselves tell Claude to execute, download or send. Line numbers (L) are valid at both commits above.
| Component | Type, size | What it does | Runs or reaches out | Verdict | Best for |
|---|---|---|---|---|---|
| seo-audit | Skill, 1 file, 384 lines | Checklist for a technical and on-page audit: crawlability, indexation, titles, content, links | Nothing | Worth installing | Anyone without an audit routine of their own |
| competitor-alternatives | Skill, 1 file, 750 lines | Structure for "X vs Y" and "X alternatives" pages in four formats | Nothing | Worth installing | Comparison and alternatives pages |
| schema-markup | Skill, 1 file, 596 lines | JSON-LD templates (Organization, Article, Product, FAQPage, Review) and validation steps | Nothing; URLs are examples inside JSON | Read first | Sites that already know which schema types they want |
| programmatic-seo | Skill, 1 file, 626 lines | Building template pages at scale from a dataset | Nothing | Read first | Ideas for data-led pages, not mass production |
| seoagent | Skill, 1 file, 101 lines | A saved SEO workflow (audit, clusters, briefs, drafts) written to a .seoagent/ folder | Web fetch of your pages (L40); points to a global npm package (L95) and a plugin marketplace (L98) | Read first | Teams who want SEO notes kept in the repo |
| content-creator | Skill, 7 files | Brand-voice and on-page analysis plus content frameworks | Runs two bundled Python scripts (L50, L78), standard library only | Read first | Writers who want a second opinion on tone |
| publisher-linkedin | Command, 64 lines | Turns a blog post into a LinkedIn draft through the LinkedIn API | Bash, WebFetch, curl to the API; token kept in .env | Skip | Nobody on a site where posts need a human OK |
| ai-directory-submitter | Skill, 1 file, 43 lines | Submits a product to AI directories from a remote list | Fetches the list fresh each run (L12), fills third-party forms, sends contact details | Skip | Nobody until the list ships inside the skill |
"Worth installing" here means the file only contains instructions, nothing in it executes, and what it teaches holds up for a content site. "Read first" means the component is fine to have, but something in it, either a command or a piece of advice, should not be followed blindly. "Skip" means the risk or the overlap outweighs what it adds. These are Saazora's editorial calls for an affiliate content site; a SaaS team or an agency might weigh some of them differently, and the sections below say where.
What aitmpl.com is, and what its installer actually puts on your machine
Claude Code Templates is a GitHub repository, davila7/claude-code-templates, with a browsing site on top. The repository is MIT-licensed (copyright Daniel (San) Ávila), and the command-line tool inside it was at version 1.30.0 on both October 19 and October 23, 2026. Each component page on aitmpl.com shows the file, a short description and a copy-paste install command such as npx claude-code-templates@latest --skill business-marketing/seo-audit.
Three facts matter before you paste that command:
- It pulls the newest npm release at install time. The
@latesttag means the installer code itself can differ from the version anyone looked at last week. That is normal for npm tools, but it is a second thing to trust on top of the component. - Skills land in your project. The installer writes a skill to
.claude/skills/<name>/(the target path is set in the CLI'sindex.js, read on October 19). Claude Code then picks it up for any matching task in that project. - Licensing is per repository, not per component. In the business-marketing folder, 8 of 49 skills state a license in their own frontmatter. Of the eight reviewed here, only
content-creatordoes (MIT, author Alireza Rezvani). The rest fall under the repository's MIT license.
The catalog also moves fast. The 50 commits before October 19 went back only to October 9, mostly automated "Update component content" commits plus the maintainer's own work, and another 42 commits landed between October 19 and October 23. A component can change between the day you read it and the day you install it, which is why the last section of this review re-checks the eight files.
Because every component on this list is a plain file, the safest install is the boring one: open the file on GitHub, read it, and copy it into .claude/skills/<name>/SKILL.md yourself. You skip the npm step entirely and you know exactly which version you have.
The two worth installing: seo-audit and competitor-alternatives
seo-audit
A 384-line checklist and nothing else. It walks Claude through an audit in a sensible priority order (can Google crawl and index the page, is the site technically sound, is the page optimized, does the content deserve to rank, does the site have credibility) and asks about the site's goals and Search Console access before it starts. There is no shell command, no network call, no hook and no MCP server anywhere in the file. On a site that already has its own audit rules it will overlap them; on a site that has none, it is a reasonable first skill. Saazora's own equivalent, and how it is used on real drafts, is covered in the round-up of Claude Code skills Saazora uses for SEO.
competitor-alternatives
The most useful file of the eight for an affiliate site, with one adjustment. Its description promises four formats: singular alternative, plural alternatives, "you vs competitor" and "competitor vs competitor". Its core principles are the ones a reviewer should want: acknowledge competitor strengths, be accurate about your own limits, don't misrepresent features (L37-39), and show a last-updated date on index pages (L289). The adjustment: it is written for a vendor promoting "your product" (L8, L14, L117, L184). An affiliate or review site is not the vendor, so keep the competitor-vs-competitor format and the honesty rules, and delete the parts that tell Claude to position one product as the winner. After that edit it is plain text that improves comparison pages.
Four to read before you let them work: schema-markup, programmatic-seo, seoagent, content-creator
None of these four is malicious. Each has one or two lines that would cause trouble if Claude followed them on autopilot.
schema-markup
Installing it is harmless: the 49 web addresses a pattern scan picks up are example.com and schema.org URLs inside JSON samples, plus two validator links. Following it as written is where it goes wrong for review sites. It includes a ready FAQPage template (L234-243) with no warning that Google no longer shows FAQ rich results, and its Product example ships an AggregateRating of "4.8" from "127" reviews (L194-197). The file does say, at L381, that self-serving reviews break the guidelines, but a model filling in a template can copy sample numbers long before it reaches that note. Use it for Organization, Article and BreadcrumbList; delete the FAQPage block and replace every sample rating before anything is published.
programmatic-seo
Plain text, and honest about its own risks: it has a section on avoiding Google penalties that warns against doorway pages, keyword stuffing and duplicate content (L78-81). The method is still the problem. Its playbooks include patterns such as "best [category]" (L112) and "[service/thing] in [location]" (L198), which is exactly the page-per-keyword fan-out that Google's scaled-content rules target. It is worth reading for ideas about data-led pages. It is not a skill to leave switched on in a project where Claude can create pages.
seoagent
The SKILL.md itself is modest: it fetches your key pages (L40), writes audit notes and briefs into a .seoagent/ folder, and recommends FAQPage and HowTo JSON-LD in places (L48, L70, L79). The part to read carefully is the end. Line 91 explains that "the full agent ships as a CLI", line 95 gives npm install -g @seoagent-official/seoagent && seoagent init, a global install of a third-party package, line 98 adds a separate plugin marketplace, and line 101 mentions an optional cloud with Search Console access and CMS publishing. Saazora did not download that package or look at the cloud service, so this review says nothing about whether they are safe. The point is narrower: installing this skill is fine, but running L95 or L98 is a separate decision about a separate product, and it deserves its own review.
content-creator
Seven files: the skill, two Python scripts, three reference documents and a calendar template. The skill tells Claude to run brand_voice_analyzer.py (L50) and seo_optimizer.py (L78). Both scripts use only Python's standard library and make no network calls, so the risk is low, but this is the one item of the eight where installing it means running code someone else wrote. Read the two scripts first; they are short. It is also the only component here with its own license line (MIT).
Two to skip for now: publisher-linkedin and ai-directory-submitter
publisher-linkedin
This slash command turns a blog post into a LinkedIn post, uploads media and creates a draft through the LinkedIn API, then opens LinkedIn for review (L49-51). To do that it grants itself Read, Write, Bash, Glob and WebFetch (L2), posts "via LinkedIn API (using Bash + curl)" (L45), and keeps a client secret and access token in your project's .env file (L56-60). That is a lot of reach for a writing task, and the token sits in a file every other tool in the project can read. Its own fallback, generating the post text for manual copy-paste (L62), is something a plain prompt already does. If LinkedIn is part of your promotion, a dedicated scheduler with drafts and approvals is the cleaner route; Saazora compares those in its LinkedIn tools hub.
ai-directory-submitter
The shortest file here, and the clearest skip. Line 12 tells Claude to download a directory list from falcoscan.com and to fetch it "fresh for each new product", so the instructions that steer the agent live on someone else's server and can change after you read the skill. It collects the founder's name, contact email and social links (L22) and submits them to third-party sites (L30). Its last instruction (L43) has Claude pitch FalcoScan's full list "for $24 once", with a tracking link tagged utm_source=skill. To be fair, it has a good stop-and-ask rule before any payment, login, account creation or CAPTCHA (L32). Saazora's guide to vetting a Claude skill before installing it uses this same file to show why remote instructions are a red flag, so the detail is not repeated here.
Does the "Scanned by NVIDIA SkillSpector" badge replace reading the files?
No, and the eight components show why. Each aitmpl.com component page now carries a "Scanned by NVIDIA SkillSpector" panel (the repository's history shows its current design landing on October 21, 2026). On October 23 every one of the eight pages showed a scan dated October 17, 2026, which the panel describes as "automated static analysis (no LLM)" that looks for "shell commands, credential access, network calls or unpinned dependencies". The panel is careful to say an observation "is something worth reading in the files before you install, not a confirmed vulnerability". Read that way, it is a useful hint. Read as a safety grade, it would mislead.
| Component | Badge on aitmpl.com | Saazora verdict | Gap |
|---|---|---|---|
| seo-audit, competitor-alternatives, programmatic-seo | 0 observations, score 0/100 | Install, install, read first | None on code. programmatic-seo's risk is the advice, which no code scanner is built to judge. |
| content-creator | 1 observation (no declared tool scope; file reading detected), score 7 | Read first | Small. The badge points at the scripts, which is the right place to look. |
| schema-markup | 1 observation at SKILL.md L542, score 16, partial analysis | Read first | L542 is a React dangerouslySetInnerHTML example for injecting JSON-LD. The FAQPage template and the sample ratings are not flagged. |
| seoagent | 0 observations, score 0, partial analysis | Read first | The global npm install (L95) and the marketplace add (L98) are not flagged. |
| publisher-linkedin | 1 observation, credential access at L46, score 15 | Skip | Flags the .env credentials, which is right; the Bash + curl posting at L45 is the bigger reason to skip. |
| ai-directory-submitter | 0 observations, score 0, partial analysis | Skip | The remote list fetched every run (L12), the personal data sent out (L22) and the paid upsell (L43) are not flagged. |
Two of the three components with a "partial analysis" note scored 0, and both are on Saazora's caution list. The practical rule: treat any observation as a pointer to a line worth reading, and treat a clean badge as "no code pattern matched", not "safe to install". Instructions written in plain English, such as "fetch this list fresh every time" or "offer the paid version", are exactly what a pattern scanner does not read as risky and a person does.
Have the eight files changed since they were read?
Not as of October 23, 2026. On October 19 the repository's head commit was 161d491 (dated October 18). On the morning of October 23 git ls-remote returned a new head, 7d85535, with 42 commits in between, among them a frontmatter fix across 253 commands and a SkillSpector report for non-skill components changed in pull requests. A fresh shallow clone into a new, empty folder, with no checkout and nothing executed, showed that none of the eight component paths had changed: git diff between the two commits printed nothing, and the folder and file hashes match exactly.
A second re-check on October 24, 2026 found head 72e5f23, six commits later. One of them (pull request #1263, "replace display-title names with the directory name") touched one of the eight files: the first frontmatter line of the seoagent skill changed from name: SEOAgent to name: seoagent. Nothing else in the eight components changed, so every quoted line and line number in this review still holds.
That is good news for this review and a reminder for anyone using it later. A folder hash is a cheap way to know whether the text you approved is still the text you are about to install. If you read a component today and install it next month, compare the commit, or simply re-open the file. For the broader set of checks Saazora runs before adding any skill to a live site, see how Saazora runs Claude Code on a live affiliate site.
What this review does not cover
A Claude Code Templates review that stays honest has limits, and these are the ones that apply here.
- The other 1,900-plus components. This is not a catalog listing. The eight were chosen from the business-marketing skills, one marketing command and a quick pattern scan of 15 candidates, because they are what a marketer would search for.
- Runtime behaviour. Nothing was installed or run, so this page describes what the files tell Claude to do, not what happened on a machine.
- The seoagent npm package and cloud, and FalcoScan's paid list. Neither was downloaded or bought.
- Popularity. Star counts and download figures change daily and are not used to rank anything here.
How this review was done
On October 19, 2026, between about 01:55 and 02:10 UTC, Saazora cloned davila7/claude-code-templates at commit 161d491 into an empty scratch folder and treated it as untrusted: files were opened and read, never installed, imported or executed. A short pattern scan (shell commands, network calls, hooks, MCP servers, home-folder paths, key names), run from outside the clone, narrowed 15 candidates to eight. Every file of those eight was then read by hand, and the lines quoted above were copied straight from the files with their line numbers. Three of the eight (seo-audit, content-creator, ai-directory-submitter) had already been read for Saazora's skill-vetting guide at commit 4df7973 on October 18 and were identical at 161d491. On October 23 the head commit was checked again, the eight paths were compared between the two commits, and the eight aitmpl.com component pages were read for their install commands and scanner results. Verdicts reflect Saazora's own publishing rules, such as no FAQPage markup and no ratings without a real source. AI tools helped draft this page; every line number, count and date was checked by hand against the saved files. No vendor saw or paid for it. More on how Saazora reviews software.
Sources (read October 19 and 23, 2026): davila7/claude-code-templates on GitHub · aitmpl.com homepage and the eight component pages.
