Marketing

Your repository includes prepackaged guidance for your own AI tool to help build and market your product. You supply the tool, account and model access; CoreMVP does not provide a managed team. Describe the audience and what you want the communication to accomplish. Lead recommends a useful direction, Code produces the authorized work, and independent Review assesses the artifact and evidence. You do not select or manage specialists.

Open the purchased repository in Codex or Claude Code. The example uses Codex $marketing; use /marketing in Claude Code. This is an example request, not a recorded delivery or evidence of marketing performance:

Use $marketing to explain our shipped product to the audience in this brief.
Recommend one useful asset and why, then create a reviewed local draft.
Use the supplied product facts and feedback. Ask me about material missing
facts or direction; do not publish or contact anyone.

Supplied facts can be your current request, existing product source, a brief or authorized feedback. The template's default brand, pricing and sample data do not establish your customers or business. No app runtime or provider setup is required merely to load this expertise.

How your AI tool uses the guidance

1. Lead reads the audience, goal, constraints, product truth and supplied evidence. Marketing identifies the communication problem, alternative and strongest objection before choosing an argument its proof can carry.
2. Lead recommends a useful asset and proportionate effort. A material question includes an executive recommendation, why, tradeoff and one decision for you. A small copy change can proceed from established facts without a strategy form.
3. Code produces and integrates the authorized draft using relevant expertise. Taste helps consequential voice, angle, composition or sequence choices; Frontend owns application visual planning when a UI change is actually requested.
4. Independent Review reads the brief, actual artifact/source and Code's evidence, returns specific findings and one quality verdict. Code fixes accepted findings; Review checks closure. Review never runs tests, operates providers or writes files.
5. Lead delivers the requested artifact with decisions, checks, limits and exact local or published state. It keeps one ignored progress record for interruptions. Distribution and later audience response retain their own boundaries.

This is a practical sequence through existing roles, not another workflow installation. Ordinary marketing does not run Development's four engineering procedures. A required runtime, metadata or other technical change uses Development under its implementation authority.

Use the expertise that fits the job

You can request a narrow task directly: Codex $writing, $taste or $seo; Claude Code /writing, /taste or /seo. These are client skill invocations, not shell or provider commands. If skill selection is unavailable, or a listed skill's description differs from this repository's copy because a same-named skill was exposed instead, explicitly load the corresponding .agents/skills/<name>/SKILL.md from this repository by path.

A short founder observation may need one social post. A reproducible technical decision may justify an article. A developer who needs to use or extend the product needs documentation. A demo needs an approved real scenario and production tools. A campaign connects distinct asset jobs; it does not require every format.

Taste grows with your product

Taste is more than a fixed visual style. It helps compare meaning, clarity, character, surprise and craft for your audience and artifact. References must be inspected; a link or a borrowed adjective is not a visual study. Frontend remains the canonical owner for unsettled application design and actual agreement.

The package starts with no observed project lessons. When genuine scoped feedback could change later work, Lead applies Taste without needing another skill request. Meaningful approvals, rejections and revisions can inform a scoped lesson; routine fixes or LGTM alone need not. Lead curates the lesson and Code persists authorized changes to the compact project-judgment support using the existing feedback method. This is explicit repository guidance, not background collection or model training. Later recommendations retrieve relevant lessons and explain where they fit. Contrary evidence can revise or remove a lesson. Your current direction and product truth take priority; another founder's preferences do not become your house style.

AI interpretations, an exported file and hypothetical customers do not establish human response. Work can finish while audience feedback is pending.

Search work begins with a useful answer

SEO decides the reader's intent and existing canonical owner before proposing another page. Improve an existing answer when it can do the job. A distinct page needs a distinct reader task and evidence, rather than a synonym or keyword quota.

The shipped application has site configuration, public/no-index metadata helpers, sitemap/robots owners and Docs/Blog metadata and content schemas. AGENTS.md points agents to these real files for the portable SEO skill. Mechanical metadata checks do not establish crawling, indexing, ranking or conversion, and robots directives do not protect private data. No analytics vendor or ongoing monitor is required by the skill.

Included guidance

Ten reusable skills and four engineering procedures are included in this checkout alongside five application skills: backend, data, auth, billing and local development. AGENTS.md supplies this foundation's architecture, permissions, component choices and SEO file map; generic skills do not replace it.

When updating, preserve your instructions, customized files and learned Taste judgment.

Client setup and finish

Canonical .agents/skills contents follow Agent Skills. Codex uses $marketing, native .codex/agents definitions and repo skill discovery. Claude Code uses /marketing, the relative CLAUDE.md -> AGENTS.md symlink, individual .claude/skills symlinks and thin .claude/agents Code/Review definitions that preload the same role skills. Lead stays in your main conversation; no native Marketing, Taste, writer or Lead subagent is needed. Code inherits normal tools and permissions; Claude Review exposes only Read/Grep/Glob. Models inherit from the parent and actual client concurrency/depth limits remain in effect.

Follow Codex Skills, Codex Subagents, Claude Skills, Claude Subagents and memory imports for client-specific details. Trust/settings, version and symlink-preserving checkout affect discovery. If native loading is unavailable, explicitly load the canonical file and assign available agents their roles. Without delegation, sequential roles are self-review, not independent review. Other standards clients need their own native integration; portable contents alone do not establish orchestration. Configuration checks do not prove automatic model matching or finished creative delivery. There is no CoreMVP-hosted service or guarantee of creative quality, growth or rankings.

Code alone executes applicable checks and records what ran, source/data used and unverified boundaries. Review is always read-only. A publish quality verdict means the artifact meets its assigned standard; it grants no distribution permission. Local drafts, repository commits, remote publication and deployment are separate. Posting, scheduling, outreach, paid calls, production and account/provider actions need explicit authority. Never put private feedback, secrets or proof credentials in a public draft or learned-judgment file.

Related guides