flaqai/backlink_skills: Codex Skills for Directory Submission and SEO Writing

The problem flaqai/backlink_skills is aimed at

Directory submission is tedious in a specific way. The work itself is not hard, but it is long, repetitive, and full of interruptions: a login wall here, a CAPTCHA there, an email confirmation that arrives twenty minutes later. Most tooling responds by automating harder, which is exactly where accounts get banned and where submissions get filed against the wrong product identity.

This repository takes the opposite position. The README states plainly that it is not a one-click mass-submission tool and that it does not promise indexing, Dofollow links, traffic or rankings. What it ships instead is a set of Codex skills that keep a human in the loop at every point where a site demands one, and that write down what happened at every site so the next run does not repeat work.

It is written for people promoting a real product: indie founders, small SEO teams, and growth engineers who already know which directories matter to them and want the execution to be traceable. The primary documentation is in Simplified Chinese, with translated README files for English, Japanese, Korean, Spanish, French, German, Russian and others.

Two submission skills, and why they are not interchangeable

The repository separates submission into two skills with genuinely different priorities.

submit-product-directories-v1-batch is built for throughput on a list you have already vetted. According to the README it normalizes URLs, strips tracking parameters, deduplicates, assigns stable queue IDs and idempotency keys, and shards the work. It performs a read-only pass first so that login walls, CAPTCHA, Turnstile and email verification surface before any form is filled. Human-required steps are collected into a single queue rather than blocking the entire batch at the first CAPTCHA. It then submits site by site in the same browser profile, writing the result before advancing the cursor, which is what makes resumption after an interruption possible.

submit-product-directories-v2-quality runs in batches of at most ten sites and front-loads research: real audience, topical relevance, editorial governance, whether the site sells links, whether it demands reciprocal links, and terms compatibility. Low-quality directory networks, rank-selling sites and forced-link schemes are filtered out before execution. Authorization is checked separately for registration, form filling, agreeing to terms, uploading assets and final submission.

The README's own guidance is to start with V2 on five to ten sites and only move vetted URLs into V1 afterwards. That ordering is the sensible part of the design: V1's speed is only worth having once you trust the list.

Installing it and running a first submission

There is no package to install. Codex Skills are directories of instructions, references and scripts, and the README says you can either clone the repository and let Codex read the relevant directory, or copy just the skill directory you need into a location Codex can discover. The README warns explicitly not to treat the whole repository as one skill, because the two submission versions and the four writing entries each have their own SKILL.md.

Clone the repository first:

git clone https://github.com/flaqai/backlink_skills.git
cd backlink_skills

Before invoking anything, prepare the inputs the README lists: canonical product name and URL, short, medium and long product descriptions, verified facts such as category, pricing and launch date, uploadable assets, and an explicit statement of which actions may run automatically and which must be asked about first. The README also says not to write passwords, OTP codes, cookies or recovery codes into task records.

Then prompt Codex with the skill name. The README gives this example for the quality-first workflow:

使用 $submit-product-directories-v2-quality，从 Free-backlink-list.md 中筛选
与我们的 AI 视频产品最相关的候选渠道。先选不超过 10 个站点，完成质量门槛、
重复检查和授权检查，再按顺序提交。低质量、强制互链、付费买链接或仅宣传排名
价值的站点直接排除。保存可审计记录，不要自动发布文章或社区帖子。

What you should see is a research and eligibility pass over the candidate sites before any form is touched, then one site at a time with a recorded outcome. If a site throws a CAPTCHA or an email verification, the skill hands it back rather than attempting to solve it.

The channel list is a candidate pool, not a free-link inventory

Free-backlink-list.md contains 743 websites or submission entry points, each with a Chinese description classifying it as a product directory, AI tool directory, startup community, software review site, content platform, business directory or plain form. The README states that historical submission status was removed so that one product's results are not mistaken for another's, and that old traffic numbers without a source or month were deleted.

The naming is the trap. The README says the file keeps the team's internal "Free Backlink List" label but that not every entry should be read as currently free to submit. The list deliberately includes sites that have shut down, entries that are closed, channels that charge, channels that require a reciprocal link, channels only suitable for article posting, duplicates, and entries that have not been verified. The README's recommendation is to have Codex re-check each site rather than submitting straight down the table.

That is an honest framing, and it is also the biggest practical cost of using this repository. You are not buying a vetted list. You are buying a starting pool plus a workflow for vetting it yourself.

Status vocabulary and the actions the skills refuse

Both submission skills use the same set of outcome labels, and the distinctions matter more than they first appear. submitted means there is a reliable receipt or a sent email. awaiting email verification and awaiting approval mean the ball is in someone else's court. published requires a public, non-preview page showing the correct product identity. submission outcome unknown means a final action was performed but the receipt is ambiguous, and the README says not to retry before investigating. submission failed requires a clear rejection, bounce or reliable failure evidence. ineligible and unavailable cover products that do not fit and sites that cannot currently be used.

The refusal list is equally explicit. CAPTCHA, Turnstile, 2FA, passkeys and email or phone verification must be completed through the site's own flow or by the user. The skills do not use CAPTCHA-solving services, stealth browsing, proxy rotation or fingerprint evasion.

The consequence is throughput that depends on your attention. A batch of sites with verification gates will stall until you clear the queue. That is the intended behaviour, not a defect, but it means the tool is a poor fit for anyone who wanted unattended overnight submission.

The four writing skills and what they actually change

The writer directory holds a general SEO writer plus three platform-specific variants. The general writer covers search-intent-driven article design, how-to and comparison and listicle formats, FAQ and meta description generation, fact-checking of product claims, prices, dates and statistics, a structured SEO audit followed by rewriting, and a humanization pass to reduce template phrasing. It plans or generates 16:9 images using local relative paths by default and writes output to writer/output/<article-slug>/. Uploading images to Cloudflare R2 is optional and the README says the workflow delivers completely without it.

The three variants are not just different output folders. linkedin-writer targets B2B readers with topic research across LinkedIn and Google-to-LinkedIn queries, business-level insight, discussion design, and a choice between Article and newsletter formats. medium-writer targets Medium's topic and publication ecosystem with author experience, narrative continuity, publication matching and AI-assistance disclosure. wechat-writer targets Chinese mobile reading with title and summary work, a claims-to-evidence ledger, phone-friendly paragraphing and a review step.

The README is clear that these skills produce local drafts or publishing packages. They do not log in and publish, and external publishing, account operations and image upload remain separately authorized actions.

Licence, maintenance and what upgrades cost you

The repository is MIT licensed, which permits commercial use, modification and redistribution provided the copyright notice and permission notice are retained. That is the standard permissive arrangement and it matters here because teams will want to copy individual skill directories into their own Codex setups rather than depending on the upstream repository. Nothing in the README adds terms on top of MIT, but the channel list is a document rather than code, and its accuracy is not warrantied by the licence either way. This is not legal advice; check the LICENSE file and your own obligations.

The last push to the default branch was on 2026-08-24, which is recent enough that the repository is not dormant. There are no releases in the retrieved data, so there is no versioned upgrade path: you track main. The practical upgrade cost is low because skills are prompt-and-reference directories rather than a running service, but it is not zero. If you copied skill directories into your own tree, upstream changes to SKILL.md will not reach you automatically.

The larger recurring cost is the channel list. Sites shut down, start charging, or change their submission rules, and the README says the list already contains entries in each of those states. Re-verifying before each campaign is part of the workflow, not an optional extra.

Where this is the wrong tool

If your goal is link volume, this repository will frustrate you. It caps the quality-first workflow at ten sites per batch, it requires authorization at several points per site, and it refuses to touch CAPTCHA or verification. Any tool that simply posts to hundreds of directories faster will beat it on count.

The second mismatch is measurement. The README says results should be judged on qualified publication rate, referral visits, conversions, profile accuracy and survival time rather than submission or link counts. If your reporting is built on link counts, this workflow produces numbers that do not fit your dashboard.

Third, the documentation is primarily Chinese. Translated READMEs exist for many languages, but if your team needs English-language reference material for the skill internals, the main SKILL.md files are what you will be reading, and the README does not describe them as translated. A team without Chinese reading ability should check the English README and the skill files before committing.

Finally, if you do not already know which directories are relevant to your product, this is not a discovery tool. The 743-entry list is a candidate pool with known gaps and dead entries, and the quality workflow expects you to bring a shortlist.

How it compares to a backlink automation tool

The obvious alternative is a hosted backlink submission service, which typically works by holding your credentials, running submissions on your behalf, and reporting a count of links created. The difference in approach is structural rather than a matter of degree.

A hosted service centralizes execution and hides the per-site decisions. This repository pushes those decisions to you and makes them explicit: which sites qualify, who authorized the submission, what evidence exists, and what the outcome label is. It runs inside Codex, so there is no vendor account and no credential handoff beyond the browser profile Codex uses.

The trade-off is real. You get auditability and a refusal to violate site rules, and you pay for it with manual verification handling and a hard ceiling on batch size. A team that wants a number in a dashboard will find the hosted route faster. A team that has already been burned by submissions attributed to the wrong product, or by accounts flagged for automated behavior, is the audience this repository was written for.

Editorial conclusion

Adopt it if you already have a verified product profile, a shortlist of directory URLs, and a Codex environment where you can keep SKILL.md files separate. Skip it if you want automated link volume or if your only goal is ranking movement, because the project states it does not promise indexing, Dofollow, traffic or rankings. Before running anything, verify two things: that the channel you picked is still live and free, since the list explicitly contains closed, paid and reciprocal-link entries, and that your task record contains no passwords, OTP codes, cookies or recovery codes.

Frequently asked questions

Official sources

1. flaqai/backlink_skills on GitHub
2. Issues
3. License: MIT
4. Project website
5. README