Codex First Customer Finder Skill: evidence-backed prospect shortlists from a startup URL
A Codex skill that finds evidence-backed potential first customers from recent public signals.
At a glance
- What is it?
- A Codex skill that turns a product URL or idea into a ranked shortlist of potential first customers, with a public source linked to every prospect and outreach left entirely manual.
- Who is it for?
- Adopt it if you are a founder or early growth person who wants a documented, source-linked shortlist to work through by hand, and you accept that every prospect is a hypothesis rather than a confirmed buyer. Do not adopt it if you need verified contact details, automated sequencing, or a CRM-ready pipeline; the README rules out private contact enrichment, and the skill explicitly does not send anything.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 1 day ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 19, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The problem: first customers are found by hand, badly
Early founders usually have a product and no list. The usual substitute is a spreadsheet of company names pulled from a directory, which says nothing about whether anyone on that list has a live problem this week. This skill targets that gap. It takes a startup URL, repository, or product description and returns a shortlist of people and companies that have recently said something public about the problem the product addresses.
The intended user is a solo founder or a small team doing pre-launch discovery, not a sales development team running outbound at volume. The README frames the output as hypotheses: prospects are candidates based on public signals, not confirmed customers or guaranteed buyers. That framing matters, because it tells you the deliverable is a research artifact you still have to act on.
How the skill turns public signals into a ranked shortlist
The pipeline has four visible stages. First the skill analyzes the input (a URL, a repository, or a written product description) and defines a primary ideal customer profile plus adjacent profiles. Second it searches public sources for five named signal types: explicit demand, pain, workaround, switching, and timing. Third it scores each prospect on fit and timing and links every primary prospect back to the original public source, with the signal date attached. Fourth it drafts a source-based outreach opener per prospect and renders everything into a standalone responsive HTML report.
Two design choices stand out. The evidence link is the core of the product: a prospect without a public source does not belong on the primary list. And the report is a single HTML file, which is why the README describes it as standalone and responsive rather than as a dashboard or a database. The scoring is the weakest documented part. The README says prospects are qualified with an evidence-based score, but it does not publish the weighting, so you cannot audit why one prospect ranked above another. Treat the ranking as a reading order, not a measurement.
Installing the Codex skill and running a first search
The README gives an npx installer as the primary path. It places the skill at ~/.codex/skills/first-customer-finder, and the README says to restart Codex afterwards. Restarting is not optional in practice: the skill is discovered at startup.
npx --yes codex-first-customer-finder-skill@latestAfter the restart, the skill is invoked by name in a Codex prompt. The README's first example asks for ten prospects and the final HTML report.
Use $first-customer-finder to find ten evidence-backed potential first customers for https://example.com and create the final HTML report.You should get back the report structure the README lists: an early-customer verdict, the primary ICP with disqualifiers, the highest-confidence prospect, the evidence-backed shortlist, fit and timing scores, source links with signal dates, outreach openers, repeated pain patterns, a seven-day manual outreach plan, and a research limitations section. If you prefer not to use npx, the README also documents a manual install that clones the repository and copies the first-customer-finder directory into ~/.codex/skills.
git clone https://github.com/Kappaemme-git/codex-first-customer-finder-skill.git
mkdir -p ~/.codex/skills
cp -R codex-first-customer-finder-skill/first-customer-finder ~/.codex/skills/first-customer-finderChoosing a mode: quick, standard, deep, design-partners, b2b, community
Mode selection changes the size and character of the output rather than the method. quick returns up to five strong prospects; standard up to ten across several source types; deep up to twenty plus repeated-pattern analysis. The remaining three are shaped by who you want to reach. design-partners prioritizes people publicly describing the problem who are likely to give product feedback. b2b looks for public business triggers and qualifies companies rather than individuals. community targets explicit requests and public discussion signals.
The practical constraint is that deep mode is the only one that adds repeated-pattern analysis, so if your goal is to learn how the problem is described in the wild rather than to build a contact list, deep is the mode that produces that. The README gives a b2b example that asks for public business triggers, qualified companies, and one opener per prospect, with the instruction not to send anything.
Where the skill stops: no contact data, no sending, no audit trail
The README is explicit that the skill avoids private contact enrichment and sensitive personal data, and that all outreach stays manual by default. That is a deliberate boundary, and it is also the main reason the tool will disappoint anyone expecting a prospecting platform. There is no email address, no phone number, no verified identity behind a handle, and no send button. You get a public source and a drafted opener, and you do the rest.
The second limitation is freshness. The whole method rests on recent public signals, and the report is a static HTML file. Nothing in the README describes a refresh, a scheduled re-run, or a diff between two reports, so a shortlist ages the moment it is written. The third is source bias: people who complain publicly about a problem are not a random sample of the market, and the disqualifiers section in the report is the only guard against that skew. The README does not document rollback or uninstall steps either, so removing the skill means deleting the directory under ~/.codex/skills yourself.
How it differs from general-purpose research agents and CRM prospecting tools
A general coding agent with web access can be asked the same question, and it will produce a list. The difference is the constraint the skill imposes: every primary prospect carries a link to the original public source and a signal date, and the output is a fixed report structure rather than free-form chat. That makes the result reviewable by someone other than the person who ran it.
Against CRM prospecting tools the difference runs the other way. Those tools start from a contact database and filter it; this skill starts from public conversation and works backwards to people. It has no database, no enrichment, and no sequence automation, so it cannot replace a sales engagement platform. It is closer in spirit to a structured literature review than to a lead list, and the seven-day manual outreach plan in the report is the handoff point where the tooling ends.
Maintenance, licence, and what you are actually installing
The package is version 0.1.0, published to npm under the name codex-first-customer-finder-skill, with a bin entry pointing at scripts/install.js. The installed payload is the first-customer-finder directory plus the installer, README, and licence. Because the npx command pins @latest, every install pulls whatever is current at that moment; if you want a fixed version you have to name it yourself, and the README does not discuss version pinning.
The repository is MIT licensed, which permits commercial use and modification provided the copyright notice and permission notice are retained. That is the licence text, not legal advice, and if you redistribute the skill inside a product you should read the LICENSE file in the repository rather than this summary. Maintenance is hard to judge from the outside: the last push to the default branch was on 2026-08-24, and no releases have been published. There is no changelog in the repository listing, so upgrades are effectively silent. The main upgrade cost is not the code but your prompts and any report templates you have come to rely on, since a change to the report structure would invalidate both.
Editorial conclusion
Adopt it if you are a founder or early growth person who wants a documented, source-linked shortlist to work through by hand, and you accept that every prospect is a hypothesis rather than a confirmed buyer. Do not adopt it if you need verified contact details, automated sequencing, or a CRM-ready pipeline; the README rules out private contact enrichment, and the skill explicitly does not send anything. Before relying on a report, verify the signal dates on the source links, because the ranking is only as current as the public posts it cites.
Frequently asked questions
How do Codex skills work?
A skill is installed into ~/.codex/skills and invoked by name in a Codex prompt, for example $first-customer-finder. This one analyzes a startup URL or product description, defines ideal customer profiles, searches public sources for demand, pain, workaround, switching and timing signals, then scores and links each prospect.
What skills should I add to Codex?
The repository does not rank skills or recommend a set. It only describes this one, which is aimed at early customer discovery: it produces a source-linked prospect shortlist and an HTML report, and it does not send outreach.
Which Codex frontend skill is the best?
The README does not compare this skill with frontend skills or rank any skill. This project is a customer discovery skill: it takes a startup URL or product description and returns an evidence-backed shortlist of potential first customers, not interface code.
What are the best design skills for Codex?
The repository does not evaluate design skills. Its own output is a standalone responsive HTML report, and the README describes only that artifact, so it offers no basis for comparing design-oriented skills.
Community notes