Affitor affiliate-skills: 52 Markdown Skills That Drive an AI Agent Through an Affiliate Funnel

The Problem: Affiliate Research Is Done on Vibes

The README states the problem bluntly in a comparison table. Without the project, it says, you Google "best affiliate programs" and get SEO spam written to rank rather than to help, you write from gut feeling with no idea what format performs, and you pick programs by vibes instead of data. That is a fair description of how most solo affiliate work actually happens. The target user is one person or a very small team running content-led affiliate promotion, not an agency with a media buying desk. The repository assumes you can operate an AI agent that reads text files, and it assumes you are willing to let that agent make program selection and angle ranking decisions from retrieved data. If your workflow already includes a paid program database or an SEO suite, the value proposition narrows considerably, because the skills are mostly orchestration and prompt structure rather than proprietary data.

How the Flywheel Chaining Actually Works

The repository layout puts every skill in skills/{stage}/{skill-name}/SKILL.md, with shared doctrine and compliance material under shared/references/ and a machine-readable catalog at registry.json. The README describes a closed loop across eight stages: S1 research, S2 content, S3 blog and SEO, S4 offers and landing, S5 distribution, S6 analytics, S7 automation and scale, with S8 meta running across all of them. The mechanism that makes this more than a folder of prompts is the chaining contract. According to the README, every skill knows what comes next through chain_metadata.suggested_next and what feeds it, and S6 analytics feeds back into S1 research. That means the agent is not choosing a tool from a list each turn; it is following declared edges in a graph. The second mechanism is research grounding. The README states that content skills do not write from thin air and that content-research-brief collects real source articles first, so posts carry real stats and quotes. The demo shows this as seven sources collected with five fully fetched, then three generated angles. Whether the agent honours the chain depends on the host, since a SKILL.md file is a text contract, not an enforced pipeline.

Install Paths and the affiliate-check CLI

The README gives four install routes. For Claude Code and Pi it recommends npx skills add Affitor/affiliate-skills. A manual clone goes to ~/.claude/skills/affiliate-skills followed by ./setup inside that directory. OpenClaw and ClawHub users run clawhub install affiliate-skills. Cursor and Windsurf use the same npx skills add command. The repository also ships tools/src/, described as the source of an affiliate-check CLI, though the README does not document its flags or output, so treat that as something to inspect in the source rather than something you can invoke from the documentation alone. There is also a zero-install path: paste a prompt into any AI that asks it to query https://openaffiliate.dev/api/programs?q=AI+video&sort=relevance&limit=5&utm_source=affiliate-skills and render the results as a table of name, reward value, cookie days, and stars. That prompt is the clearest statement of the project's external dependency, and it is also the fastest way to judge whether the data is good enough for your niche before you install anything.

The Demo Numbers Are Illustrative, Not Measured

The README's zero-to-first-commission walkthrough contains a lot of specific figures: 47 videos scanned, a comparison format at 45 percent of top content with average engagement 35.2, a best hook at 42.3, a median of 18K views with the top decile at 85K or more, and a program table listing HeyGen at 30 percent recurring with a 60 day cookie and a traffic score of 82 out of 100. None of these are presented as reproducible benchmark results, and the repository has no published release notes to check them against. They are a scripted illustration of what the output shape looks like. That distinction matters when you evaluate the project, because the honest question is not whether 35.2 is the right engagement number but whether trending-content-scout can produce that shape of output for your topic at all. The same caution applies to the traffic analyzer's website health score, which the README describes as due diligence on advertiser site health before you commit to promoting. A composite score of that kind is only as good as its inputs, and the README does not enumerate them.

Where the Design Breaks Down

The first limitation is the data dependency. Program search, commission figures, cookie windows, and traffic scores all come from openaffiliate.dev. If that API does not cover your vertical, or if it changes shape, the S1 research stage degrades and every downstream stage inherits the gap. The second limitation is that the content and distribution skills are instructions to an agent, not integrations. The README lists pages, bio links, email sequences, and deployment under the builder stage, but nothing in the supplied material describes a deployment target, a hosting integration, or credentials handling, so you should expect the agent to improvise with whatever tools it already has. The third is the compliance surface. The README mentions an FTC audit skill and shows an FTC disclosure in the generated LinkedIn post, which is the right instinct, but disclosure rules vary by jurisdiction and platform, and a skill that audits your draft is not a substitute for knowing the rules that apply to you. Finally, the primary language of the repository is HTML, which fits the infographic and landing page output but tells you little about the surrounding tooling.

How It Differs From a Plain Prompt Library

The obvious alternative is a folder of your own prompts, or one of the many prompt collections that circulate for marketing tasks. The difference here is the declared graph. A prompt library gives you independent tools and leaves sequencing to you. affiliate-skills encodes the sequence in registry.json and in per-skill chain_metadata.suggested_next, so an agent that reads those files can move from scout to brief to angle ranker to writer without you restating context each time. The second difference is the shared references directory, which holds doctrine, compliance, and flywheel material that individual skills can cite instead of repeating. A third option is a dedicated affiliate network dashboard, which gives you accurate commission data but no content generation and no angle ranking. affiliate-skills sits between those poles: weaker data than a network dashboard, stronger sequencing than a prompt folder. Which side of that trade you prefer depends on whether your bottleneck is knowing what to write or knowing what to promote.

Maintenance Cost and Licence

The repository is MIT licensed per the README badge and the LICENSE file referenced in the structure, and it is not archived. There are no retrieved releases, so there is no changelog to diff against when you upgrade. That pushes the upgrade cost onto you: because skills are plain Markdown under skills/{stage}/{skill-name}/SKILL.md, a git pull can silently change the instructions your agent follows, and the chaining contract in registry.json can shift underneath skills you have already tuned. The practical mitigation is to fork the repository and pull deliberately rather than tracking main. On licence, MIT is permissive, but note that the skills generate affiliate content and the repository bundles no content of its own; the disclosure and advertising rules that apply to what you publish are yours, not the licence's. Nothing here constitutes legal advice, and the compliance material under shared/references/ is documentation rather than a guarantee.

Who Should Adopt This, and What to Check First

Adopt it if you already have an agent host that reads SKILL.md files, you promote in a vertical that openaffiliate.dev covers, and you want the funnel structure handed to you rather than assembled. Do not adopt it if you need a self-contained pipeline, if you cannot tolerate an external API in the critical path of program selection, or if your compliance obligations require a review process the skills cannot express. The first thing to verify is coverage: run the README's no-install prompt against your own niche and look at whether the returned programs and cookie windows are ones you would actually promote. The second is the chain contract: read one stage's SKILL.md end to end and confirm that chain_metadata.suggested_next points where you expect. The third is the licence file in your clone, since the badge is not the text. If all three hold, the eight-stage loop is a reasonable scaffold; if the first one fails, nothing downstream will save it.

Editorial conclusion

Adopt affiliate-skills if you already run an agent that reads SKILL.md files and you want a structured eight-stage funnel rather than a pile of loose prompts. Skip it if you need a self-contained system: program data comes from openaffiliate.dev, and the content and distribution skills assume your agent has its own web and platform access. Before committing, read skills/{stage}/{skill-name}/SKILL.md for one stage end to end, check registry.json for the chain_metadata.suggested_next contract, and confirm the MIT LICENSE file is present in the clone, since the README badge alone is not the licence text.

Official sources

1. Affitor/affiliate-skills on GitHub
2. Issues
3. License: MIT
4. Project website
5. README