Skills for Product, Marketing and Knowledge Work

Product, marketing and knowledge-work skills teach a coding agent the jobs around a release: Corey Haines’s Marketing Skills (50 skills for conversion, copy and SEO), last30days for cited research on the past month, Steph Ango’s Obsidian skills for vault notes, and the gws skills for Google Workspace. All four install in Claude Code, Codex and Cursor.

The feature is merged and the tests are green. Now someone has to rewrite the pricing page, draft the launch post, check the new docs pages for SEO problems, and find out what people are saying about the competitor that shipped last week. Without a skill, the agent writes generic copy with invented numbers. This page is for developers and tech leads who ship their own launches, and for the product people who work next to them in the same repository.

What you’ll walk away with from product and marketing skills

• A shortlist of five skill collections, with install commands for each agent and their measured context cost
• A worked /pricing conversion audit: the install, the prompt, and the output you should see
• A launch checklist that chains research, copy and an SEO audit, with a verification gate at each step
• The traps that make these skills fail quietly: paid placements, cookie access, missing shared skills, and more

Which product and marketing skills are worth installing?

Five collections cover most of the non-code work around a software launch. Popularity is GitHub stars as of 2026-09-26, read from the GitHub API. Stars measure attention on a repository, not use of one skill.

Collection	Publisher	What it does	Stars (2026-09-26)	Always-on context cost
Marketing Skills	Corey Haines (community, MIT)	50 skills: cro, copywriting, seo-audit, launch, pricing, ab-testing, analytics and more, all reading one shared product context file	51.5k	~13,571 tokens for the whole plugin
last30days	Matt Van Horn (community, MIT)	One skill that searches Reddit, X, YouTube, Hacker News, GitHub, Polymarket and the web for the last 30 days and writes a cited brief	62.8k	~105 tokens; ~101.5k tokens each time it runs
Obsidian skills	Steph Ango (kepano), CEO of Obsidian, in his personal repository (MIT)	Six skills: obsidian-markdown, obsidian-bases, json-canvas, obsidian-cli, defuddle, knap	48.9k	~597 tokens
Google Workspace CLI skills	googleworkspace org; the README says “This is not an officially supported Google product”	gws-* skills that drive the gws CLI for Drive, Gmail, Calendar, Docs and Sheets	31.2k	Not measured (not a Claude Code plugin)
Wondel.ai skills	Wondel.ai, the publisher of this site	65 skills built on business books (StoryBrand, $100M Offers, Jobs to Be Done) plus 14 guided “metaskills”	not ranked (publisher of this site)	~1,073 tokens for the 5-skill marketing-cro plugin

The context costs come from claude plugin details <plugin>@<marketplace> on Claude Code 2.1.283 (the latest channel on 2026-09-26). “Always-on” is the skill descriptions the agent carries in every session; the full skill text loads only when a skill fires.

Disclosures

Marketing Skills carries paid placements. Its README has a “Verified Partners” block (Converly and Ploy) that funds the library, and links to the author’s agency and courses. The README says partners never influence “what the core skills recommend”; we have not audited that claim, so check any tool a skill recommends against a neutral alternative. Wondel.ai publishes both this site and the Wondel.ai skills, so read its row with that in mind.

How to pick a skill for the job

If you need to…	Use	Why this one
find why a pricing, landing or signup page does not convert	cro (Marketing Skills)	It audits in a fixed order (value proposition, headline, CTA, trust, objections, friction) and returns quick wins, bigger changes and test ideas
write or rewrite page copy, emails or a launch post	copywriting, copy-editing	They read your product context first, so the copy uses your positioning instead of generic claims
check a site or a new section for SEO problems	seo-audit, schema, ai-seo	A technical and on-page checklist, including hreflang for multilingual sites
learn what users and competitors said this month	last30days	It scores results by engagement and cites every source
turn meeting notes into linked notes and a task view	obsidian-markdown, obsidian-bases	They write valid wikilinks, callouts, properties and .base files
read or update Drive, Docs, Sheets or Gmail from the agent	gws-drive, gws-docs, gws-sheets, gws-gmail	The gws CLI returns structured JSON and has --dry-run
work through a framework from a business book, step by step	Wondel.ai conversion-optimization, storybrand-messaging, jobs-to-be-done	Framework skills; the conversion-optimization metaskill chains them phase by phase and keeps its state in docs/

For code-facing skills, see official skills from technology makers and best skills for frontend development. For decks, Word files and prose review, see document and writing skills. For day-to-day productivity skills, see top productivity skills, and for the whole catalogue, start at the skills hub.

How do you install the marketing and research skills?

Install at project scope for skills that read your repository (Marketing Skills, Wondel.ai), so the team shares one version through code review. Install research and personal-notes skills (last30days, Obsidian) at user scope with -g. Pick three to five marketing skills, not all 50.

Claude Code

# Marketing Skills: exactly the five skills this page uses (project scope)
npx skills add coreyhaines31/marketingskills \
  --skill product-marketing cro copywriting seo-audit launch --agent claude-code -y

# last30days as a plugin (user scope)
claude plugin marketplace add mvanhorn/last30days-skill
claude plugin install last30days@last30days-skill

# Obsidian skills as a plugin
claude plugin marketplace add kepano/obsidian-skills
claude plugin install obsidian@obsidian-skills

The whole Marketing Skills collection is also a plugin (claude plugin marketplace add coreyhaines31/marketingskills, then claude plugin install marketing-skills@marketingskills), but it adds all 50 skills and ~13,571 always-on tokens. Plugin skills are namespaced, for example /last30days:last30days; skills from the skills CLI are called by name, for example /cro.

Codex

npx skills add coreyhaines31/marketingskills \
  --skill product-marketing cro copywriting seo-audit launch --agent codex -y
npx skills add mvanhorn/last30days-skill -g --agent codex -y
npx skills add https://github.com/kepano/obsidian-skills -g --agent codex -y

The skills CLI writes Codex’s project skills to .agents/skills/ and global ones to ~/.codex/skills/. Invoke a skill with $cro or $last30days, and type /skills to see what loaded.

Cursor

npx skills add coreyhaines31/marketingskills \
  --skill product-marketing cro copywriting seo-audit launch --agent cursor -y
npx skills add mvanhorn/last30days-skill -g --agent cursor -y
npx skills add https://github.com/kepano/obsidian-skills -g --agent cursor -y

The skills CLI writes Cursor’s project skills to .agents/skills/, the same folder Codex reads. Ask for a skill by name in Agent chat (“use the cro skill”).

We ran the Marketing Skills command with --agent claude-code codex cursor (skills CLI 1.7.0): it wrote the five skills to .agents/skills/, symlinked them into .claude/skills/, and wrote skills-lock.json. Commit the lock file. The CLI ends every install with “Review skills before use; they run with full agent permissions”; skill supply-chain security has the checklist, and installing and managing skills covers updates and removal.

Install the Google Workspace skills

The gws skills call a CLI, so install the CLI and sign in first. In the terminal:

npm install -g @googleworkspace/cli        # v0.22.5 on 2026-09-26; or: brew install googleworkspace-cli
gws auth setup                             # one-time Google Cloud project and OAuth client; needs gcloud
gws auth login -s drive,sheets            # request only the services you use
npx skills add https://github.com/googleworkspace/cli \
  --skill gws-shared gws-drive gws-sheets --agent claude-code codex cursor -y

Every gws-* skill tells the agent to read gws-shared first for authentication, global flags and its security rules, so install it with the others.

How do you run a CRO audit of a /pricing page with the cro skill?

This is the full usage example. The cro skill reads a product context file, audits the page, and returns four sections: Quick Wins, High-Impact Changes, Test Ideas and Copy Alternatives. The example assumes a web app with a /pricing route and a local dev server.

1. Write the product context once. Every Marketing Skills skill reads .agents/product-marketing.md before it does anything. Without it, the agent asks you the same questions in every session or guesses. Run the prompt below, answer the questions, and commit the file.

   Copy-paste prompt: create the product context file

   Use the product-marketing skill to create .agents/product-marketing.md.
   Before asking me anything, read README.md, the pricing page source, the
   last 20 entries of CHANGELOG.md and any docs under docs/ that describe
   customers. Draft every section you can from those files and mark each
   statement with its source file. Then ask me only about what you could not
   find. Do not invent customer names, numbers, testimonials or guarantees:
   leave a TODO where the evidence is missing.

2. Run the audit against the rendered page, not only the source. The skill judges what a visitor sees. Start the dev server, then give the agent the URL and the source file.

   Copy-paste prompt: /pricing CRO audit

   Use the cro skill to audit http://localhost:3000/pricing. The source is
   app/pricing/page.tsx. Read .agents/product-marketing.md first.
   Return the skill's four sections: Quick Wins, High-Impact Changes,
   Test Ideas and Copy Alternatives. For each finding, quote the exact
   element on the page and name the file and line to change. For each
   test idea, write a one-sentence hypothesis, the primary metric, and
   the analytics event we need to track it. Do not edit any file yet.

3. Read the report, not the diff. Typical findings are quick wins such as a vague headline or two competing CTAs, structural changes such as moving objection handling next to the plan table, and a small number of A/B tests. Reject any finding that does not quote the page: the skill should be reading your page, not reciting its checklist.

4. Apply the quick wins on a branch and verify them. Ask the agent to implement only the accepted findings, then run the gates in the next section before anyone reviews the copy.

How do you verify marketing output without reading every line?

Marketing output fails in two ways: the page breaks, or the copy says something that is not true. Automate the first and give the second a single, narrow review.

• Claims trace to the context file. Run a second agent pass that lists every factual claim in the changed copy (numbers, customer names, guarantees, integrations, “SOC 2” style badges) and marks each as found in .agents/product-marketing.md or unsupported. A human reviews only the unsupported list. The prompt below does this.
• The page still works. Your existing type check, lint and end-to-end tests run on the branch. Add one Playwright assertion per changed CTA (it exists, it is visible, it links to checkout), or let the agent check the running page with agent-browser.
• SEO checks run in a browser. The seo-audit skill itself warns that web_fetch and curl cannot see JSON-LD injected by JavaScript, so “no schema found” from a fetch is not a finding. Confirm structured data in a rendered page or Google’s Rich Results Test.
• The change is measured, not argued. Every test idea ships behind the analytics event named in the audit, with the hypothesis written down before launch. The product owner signs off on the hypothesis and the unsupported-claims list; engineering signs off on the green gates.

Copy-paste prompt: claims check before review

Compare the working tree with main. For every user-facing string that
changed, list each factual claim: numbers, prices, customer or partner
names, certifications, guarantees, integrations and time promises.
For each claim, quote the line in .agents/product-marketing.md or
pricing config that supports it, or mark it UNSUPPORTED. Output a table
with file:line, claim, evidence, status. Do not change any file.

Launch checklist: research, copy and an SEO audit

This is the framework-level workflow. It runs the three skills in the order a launch needs them, with a gate between steps, so each step works from checked input rather than from the previous step’s guesses.

1. Research (last30days). Before writing anything, learn what the market says this month. In Claude Code, run /last30days followed by your category and main competitor (from the plugin it can appear as /last30days:last30days); in Codex, use $last30days. The README also documents discovery mode, for example /last30days what's exploding in AI agents?. Gate: open three cited sources at random and confirm each one says what the brief claims and falls inside the 30-day window. Save the brief next to the launch plan.

2. Plan the launch (launch). Ask the launch skill for a plan based on the brief and the product context. It includes a readiness gate (is the release Simple, Lovable and Complete?) to run before the phased launch plan. Gate: a product owner accepts or cuts the plan’s scope.

3. Write the copy (copywriting, cro). Draft the announcement, the changelog entry and the landing page changes, then run the cro audit from the previous section on the landing page. Gate: the claims-check prompt returns no UNSUPPORTED rows, and the page tests pass.

4. Audit SEO (seo-audit). Run the audit on the new or changed pages only. Gate: every critical finding is fixed or recorded as a ticket, and structured data is confirmed in a rendered page.

5. Ship and measure. Merge through your normal review and release process. Record the metric and the event from the launch plan, and review it at the date the plan names.

Copy-paste prompt: research brief to launch plan

Read docs/launch/last30days-brief.md and .agents/product-marketing.md.
Use the launch skill to plan the launch of the feature described in the
last CHANGELOG.md entry. Run the readiness gate first and tell me which
of Simple, Lovable and Complete fails, if any. Then list the owned,
rented and borrowed channels you recommend, each with one message
angle taken from the brief and the source link that supports it.
Keep the plan to one page and write it to docs/launch/plan.md.

When do Obsidian and Google Workspace skills belong in a coding agent?

Use them when the input or output of engineering work lives outside the repository. Two patterns are common.

• Obsidian vault as the team’s notes. Run the agent from the vault root. obsidian-markdown writes valid wikilinks, callouts and properties, and obsidian-bases writes .base views that Obsidian renders as tables. obsidian-cli also needs the Obsidian CLI, which comes with the desktop app. Put the vault under Git, or back it up, before the first run, so you can review and revert the agent’s edits.
• Google Workspace as the source of truth. A sprint review doc in Drive, a pricing sheet in Sheets, or a customer thread in Gmail. The gws CLI returns JSON the agent can parse, and gws-shared tells the agent to confirm before any write or delete and to prefer --dry-run.

Copy-paste prompt: Drive sprint review to Obsidian note

Use gws-drive to find the Google Doc named "Sprint review" modified in
the last 7 days and export it as plain text. Do not modify anything in
Drive. Then use obsidian-markdown to write Meetings/sprint-review.md
in this vault: a decision callout for each decision, a wikilink to each
person's note under People/, and a task list of action items with owners.
Use obsidian-bases to create Bases/open-actions.base listing every open
action item in Meetings/ grouped by owner.

Book-framework skills from Wondel.ai

Wondel.ai (the publisher of this site) maintains 65 skills based on business and engineering books. The business ones apply a named framework rather than a checklist: StoryBrand for messaging, $100M Offers for packaging, Jobs to Be Done for discovery, Nielsen’s heuristics for usability. The 14 metaskills chain several of them. conversion-optimization, for example, runs lean-analytics, cro-methodology, storybrand-messaging, hundred-million-offers, influence-psychology and design-everyday-things phase by phase, and records its state in docs/FUNNEL.md, docs/METRICS.md and docs/EXPERIMENTS.md so the work resumes across sessions.

# Individual skills, any agent
npx skills add wondelai/skills --skill cro-methodology storybrand-messaging jobs-to-be-done --agent claude-code codex cursor -y

# Claude Code: a plugin of five marketing skills (~1,073 always-on tokens)
claude plugin marketplace add wondelai/skills
claude plugin install marketing-cro@wondelai-skills

Reach for these when a team wants a shared vocabulary (everyone writes job statements the same way). Reach for Marketing Skills when you want a practitioner’s checklist applied to a specific page. The two overlap on CRO, so install one CRO skill, not both.

Traps with marketing, research and workspace skills

What fails quietly

• Installing from inside an agent session. The Marketing Skills README warns that a non-interactive npx skills add may write only to .agents/skills/, which Claude Code does not read. Pass --agent claude-code (or -a claude-code) explicitly.
• Old skill names. Marketing Skills v2.0 renamed 17 skills and merged page-cro and form-cro into cro. Guides that name page-cro, pricing-strategy or launch-strategy are out of date, and upgrading leaves the old folders beside the new ones. The skills CLI matches --skill against the name in SKILL.md, so a stale name matches nothing.
• last30days reads your browser cookies and runs scripts. Its frontmatter grants Bash, Read, Write and WebSearch; X search uses browser cookies, and some sources need API keys stored in ~/.config/last30days/.env. It needs Python 3.12 or later. Run python3 skills/last30days/scripts/last30days.py --preflight from a checkout to see what it would read and write, and keep it out of CI.
• last30days is expensive per run. claude plugin details reports ~101.5k tokens each time it fires. Run it once per launch, save the brief, and have later steps read the file.
• Two copies of one skill. The last30days README notes that Claude Code does not deduplicate a plugin install and an npx skills install; you see two /last30days entries. Use one install method per machine.
• gws without gws-shared. Each gws-* skill starts by reading ../gws-shared/SKILL.md. Without it, the agent loses the auth and safety rules.
• Too many OAuth scopes. An unverified Google OAuth app is limited to about 25 scopes, and the recommended preset has more than 85. Request only the services you need, for example gws auth login -s drive,sheets.
• “Official” labels. googleworkspace/cli is not an officially supported Google product, and the Obsidian skills live in the CEO’s personal repository, not the obsidianmd organisation.

When product and marketing skills break

The copy is generic or wrong about the product. The context file is missing or thin. Check that .agents/product-marketing.md exists (older installs used .claude/product-marketing-context.md), then rerun the context prompt with more source files.

The audit recites a checklist instead of reading the page. The agent could not load the URL. Confirm the dev server is running, or paste the rendered HTML, and require a quoted page element for every finding, as the audit prompt does.

Responses slow down or skills stop firing. Too many descriptions are loaded. Run /context in Claude Code, remove the skills you are not using this week with npx skills remove <name>, and prefer the five-skill install over the 50-skill plugin.

The research brief cites something that does not say what the brief claims. Treat the brief as leads, not facts. Remove the claim, rerun last30days with a narrower query, and spot-check again.

The agent wrote to Drive or Gmail when you asked it to read. Check the change in the document’s version history and restore it there. Then remove the app’s access in your Google Account’s third-party access settings, sign in again with only the services the task needs, and put “Do not modify anything” in every read-only prompt. The gws-shared rules ask the agent to confirm writes, so a write without a confirmation prompt means the skill was not loaded.

Where to go next with product and marketing skills

Product management when build time collapses Intent files, acceptance criteria and outcome roadmaps for product managers working with agents.

Skill supply-chain security Vet a third-party skill before it runs with your agent's permissions.

Documents and writing skills Decks, Word files and prose review for the documents around a release.

Building your own skills Turn your team's launch checklist into a skill of its own.