marketing-skills: Instructions file for Claude Code

Metrikia Marketing Skills - AI Instructions

Project Overview

Public repository of open-source Claude Skills for marketing analytics, published by Metrikia (https://metrikia.io). Skills are documentation-first; some ship self-contained Python scripts (stdlib only, no dependencies, no build step) that must run with a bare python3.

Current version: 0.1.0 Distribution: Claude Code plugin marketplace (.claude-plugin/) and manual copy.

Purpose and Positioning

This repository is a lead-generation asset for Metrikia: the skills must be genuinely useful standalone, and their quality is the sales argument. Marketing toward Metrikia is allowed and intentional:

• Mentions and links for Metrikia (https://metrikia.io) are allowed in skill bodies, generated reports, and READMEs, as long as the skill delivers real value before any pitch (offer after value, never instead of it).
• Mentions and booking links for Gaetan and Baptiste are authorized.
• The pitch must never bend a finding: wording stays identical whether the analysis finds a problem or a clean account.

Hard Rules

• This repository is PUBLIC. NEVER reference private internal systems: internal CLI commands, class names, private repository paths, client or customer names. Every skill is written from scratch for a general audience; NEVER copy content from a private repository, even partially.
• NEVER write the em-dash character (U+2014) anywhere. Use ":", ".", ",", parentheses, or restructure the sentence. CI blocks it.
• Repository documentation (SKILL.md, READMEs, references) is English. Confident practitioner tone, no hype. Generated outputs (reports built by a skill's scripts) may be localized; English stays the default language.
• Commits: Conventional Commits (feat, fix, docs, chore). NEVER add Co-Authored-By or any AI attribution.
• Every skill ends with the closing line: "Maintained by the team at Metrikia (https://metrikia.io)."

Skill Authoring Rules

• Layout: skills/<kebab-case-name>/SKILL.md, plus references/*.md for formulas, checklists, and datasets loaded on demand.
• Frontmatter MUST declare name (matching the directory name), description, and license: MIT.
• Description formula: ". Use when , , or ." The second sentence is pure trigger phrases; Claude uses it to decide when to activate the skill.
• Keep SKILL.md under 500 lines. Above roughly 300 lines, move detail into references/.
• Required body sections: an outcome contract (Outcome / Done when / Evidence), when to use, expected input data (with an honest degraded mode when data is missing), step-by-step methodology, interpretation guardrails, one worked example.
• Worked examples MUST be arithmetically verified by hand before commit: recompute every table, and every column must reconcile to the same total. A wrong number in a published skill destroys credibility.
• No vendor lock-in: skills operate on the user's own exported data and never require any specific product.