OpenDirectory: 64 Marketing Agent Skills Installed by npx
AI Agent Skills built for Founders who hate Marketing
At a glance
- What is it?
- OpenDirectory is an MIT-licensed pack of agent skills for GTM and marketing tasks, installed into Claude Code, Codex, Gemini CLI and five other agents through a single npx command. It is a prompt library, not a runtime: the value sits in the skill files, and the README's ranking of them is the author's own.
- Who is it for?
- Adopt OpenDirectory if you already run a coding agent and want marketing task scaffolding without writing your own prompt files: install one skill first, for example npx "@opendirectory.dev/skills" install human-tone --target claude, and read the installed file before trusting the output.
- 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 30 days 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 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The problem OpenDirectory targets: marketing prompts that founders never write
Founders running small teams tend to do marketing work in the same editor where they write code. The agent is already open. What is missing is the instruction layer: what a pricing page audit should check, how a cold email should be structured, which patterns make copy read as machine-written. OpenDirectory packages that layer as installable skill files. The README describes each skill as "a pre-built set of instructions and context you install directly into your AI agent." The intended user is a technical founder or a small GTM team already using Claude Code, Codex, Gemini CLI, OpenCode, Anti-Gravity, OpenClaw, Hermes or Manus AI. The name is a slight misnomer: this is not a directory service with an API, it is a curated file collection distributed through npm and skills.sh. Anyone expecting a hosted tool with a dashboard will find a CLI and a set of text files instead.
What a skill actually is here, and how the npx CLI moves it
Every skill lives in its own folder under skills/ in the repository, and the install command copies that folder into the target agent's skill location. The CLI takes two arguments: a skill name and a --target flag naming the agent. The README's table maps each agent to a flag, so --target claude, --target opencode, --target codex, --target gemini, --target anti-gravity, --target openclaw and --target hermes are the documented values. Running the package with no arguments opens an interactive TUI browser with categories, search and an install flow. Running it with list prints all skills. Nothing in the supplied material indicates a server component, a hosted index, or telemetry; the mechanism is fetch, resolve target directory, write files. That is the whole data flow, and it explains why the project needs no account and no API key. It also explains the ceiling: a skill can only change what the agent does with the context it already has, unless the skill itself shells out to something like Apify, which the twitter-GTM-find-skill description mentions.
Install commands and the Node.js requirement
The recommended path is npx, which fetches and runs the latest version without a global install. The README gives the example npx "@opendirectory.dev/skills" install show-hn-writer --target claude, and instructs replacing the skill name and the agent flag. To see what is available, npx "@opendirectory.dev/skills" list prints the catalogue; the interactive browser is the same command without arguments. There is a second distribution channel through skills.sh: npx skills add Varnan-Tech/opendirectory installs the whole pack and auto-detects the coding agent, and npx skills add Varnan-Tech/opendirectory --skill <skill-name> narrows it to one. Node.js is required, and the README points to nodejs.org for installation. One detail worth noting for anyone scripting this: because npx resolves the latest version each time, a pinned version is not the default behaviour, and the README does not describe a flag for pinning. Teams that need reproducible installs will have to handle that outside the documented commands.
The catalogue: SEO and psychology skills next to lead-scraping skills
The README's popular list mixes very different kinds of work. position-me is described as a website reviewer for AEO, GEO, SEO, UI/UX psychology and copywriting. pricing-page-psychology-audit checks a SaaS pricing page against 12 pricing psychology principles and returns a ranked report. human-tone rewrites marketing copy against what the README calls 18 GTM slop patterns, with before and after audit notes. On the data side, cold-email-verifier verifies cold emails, enriches lead lists and can guess addresses from a CSV; npm-downloads-to-leads scores npm download velocity to surface maintainers with outreach briefs; yc-intent-radar-skill scrapes daily job listings from Work at a Startup without duplicates; twitter-GTM-find-skill looks for X/Twitter GTM hiring signals through Apify. Two entries stand apart. meta-tribeV2-skill is described as analysing video hooks using Meta's TRIBE v2 fMRI model, which implies a research model rather than a marketing heuristic. blog-cover-image-cli generates a 1200x630 cover with what the README calls a self-healing QA loop that retries up to three times with vision feedback. That last one is the only entry in the list whose description includes a concrete retry bound.
Where the pack is thin: ranking, verification and dependency weight
The README states that the popular skills are "ranked by artifact quality and practical value" and that any skill installs in under 30 seconds. Both claims come from the maintainers, and no release notes were retrieved to corroborate them, so treat the ordering as editorial rather than measured. The deeper limitation is what a skill cannot do. Several listed skills depend on external services: Apify for X/Twitter signals, a vision-capable model for the cover-image QA loop, and whatever the TRIBE v2 analysis requires. The README does not document API keys, rate limits or costs for those dependencies, and the supplied material does not say how failures are surfaced when a key is absent. There is also no documented uninstall or update command, no version pinning, and no described mechanism for detecting that an installed skill has drifted from the repository copy. If your marketing work is paid acquisition, lifecycle automation or analytics instrumentation, the skill names shown here do not cover it. This is a prompt library with a scraper tier attached, not a marketing platform.
How it differs from writing your own skills or using a prompt template site
The closest alternative is the built-in skill or custom-instruction mechanism each agent already ships, where you write the prompt file yourself and keep it in your own repository. That approach costs an afternoon per skill and gives you full control over wording, but it produces nothing reusable across agents. OpenDirectory's difference is portability: one skill folder is written once and installed into eight targets through the --target flag, so a team split between Claude Code and Codex can share the same pricing-page audit. The second alternative is a prompt template marketplace, where you copy text out of a browser and paste it into a chat. That gives you no file on disk, so the agent cannot load it automatically at the start of a session, which is the mechanism the README relies on. The trade-off is the inverse of the DIY route: you get breadth and cross-agent installation, and you give up control over the exact instructions your agent follows.
Licence, maintenance and what upgrading actually costs
The repository is MIT-licensed, which permits commercial use, modification and redistribution provided the copyright notice and permission notice are retained; this is a description of the licence terms, not legal advice, and anyone embedding the skills in a product should read the LICENSE file as shipped. Maintenance cost splits in two. The first part is cheap: because npx resolves the latest package each time, upgrading means running the install command again, and the README documents no migration step or state to preserve. The second part is not cheap. Skills are instructions, so a rewrite of a skill changes agent behaviour with no compiler to catch the difference, and the README describes no changelog or versioning scheme for individual skills. The repository shows no releases in the retrieved metadata, so there is no release history to diff against. A team depending on a specific skill's wording, human-tone with its 18 patterns for instance, has no documented way to freeze that wording short of copying the folder and maintaining a fork.
Editorial conclusion
Adopt OpenDirectory if you already run a coding agent and want marketing task scaffolding without writing your own prompt files: install one skill first, for example npx "@opendirectory.dev/skills" install human-tone --target claude, and read the installed file before trusting the output. Do not adopt it if you need a service with a support contract, if your agent is not one of the eight listed targets, or if your marketing work is mostly paid acquisition, where the skill list shown in the README offers nothing. Verify the skill count and the target list against the repository itself before committing, since no release notes were retrieved.
Community notes