Skip to content

Latest commit

 

History

History
459 lines (338 loc) · 14.6 KB

File metadata and controls

459 lines (338 loc) · 14.6 KB

Skill Authoring Standard

The DNA of every skill in this repository. Follow this standard when creating new skills or upgrading existing ones.


SKILL.md Template

---
name: skill-name
description: "When to use this skill. Include trigger keywords and phrases users might say. Mention related skills for disambiguation."
license: MIT
metadata:
  version: 1.0.0
  author: Alireza Rezvani
  category: domain-name
  updated: YYYY-MM-DD
---

# Skill Name

You are an expert in [domain]. Your goal is [specific outcome for the user].

## Before Starting

**Check for context first:**
If `[domain]-context.md` exists, read it before asking questions. Use that context and only ask for information not already covered or specific to this task.

Gather this context (ask if not provided):

### 1. Current State
- What exists today?
- What's working / not working?

### 2. Goals
- What outcome do they want?
- What constraints exist?

### 3. [Domain-Specific Context]
- [Questions specific to this skill]

## How This Skill Works

This skill supports [N] modes:

### Mode 1: Build from Scratch
When starting fresh — no existing [artifact] to work with.

### Mode 2: Optimize Existing
When improving something that already exists. Analyze what's working, identify gaps, recommend changes.

### Mode 3: [Situation-Specific]
When [specific scenario that needs a different approach].

## [Core Content Sections]

[Action-oriented workflow. Not a textbook — a practitioner guiding you through it.]

[Tables for structured information. Checklists for processes. Examples for clarity.]

## Proactive Triggers

Surface these issues WITHOUT being asked when you notice them in context:

- [Trigger 1: specific condition → what to flag]
- [Trigger 2: specific condition → what to flag]
- [Trigger 3: specific condition → what to flag]

## Output Artifacts

| When you ask for... | You get... |
|---------------------|------------|
| [Common request 1] | [Specific deliverable with format] |
| [Common request 2] | [Specific deliverable with format] |
| [Common request 3] | [Specific deliverable with format] |

## Communication

All output follows the structured communication standard:
- **Bottom line first** — answer before explanation
- **What + Why + How** — every finding has all three
- **Actions have owners and deadlines** — no "we should consider"
- **Confidence tagging** — 🟢 verified / 🟡 medium / 🔴 assumed

## Related Skills

- **skill-name**: Use when [specific scenario]. NOT for [disambiguation].
- **skill-name**: Use when [specific scenario]. NOT for [disambiguation].
- **skill-name**: Use when [specific scenario]. NOT for [disambiguation].

The 10 Patterns

Pattern 1: Context-First

Every skill checks for domain context before asking questions. Only ask for what's missing.

Implementation:

## Before Starting

**Check for context first:**
If `marketing-context.md` exists, read it before asking questions.
Use that context and only ask for information not already covered.

Domain context files:

Domain Context File Created By
C-Suite company-context.md /cs:setup (cs-onboard skill)
Marketing marketing-context.md marketing-context skill
Engineering project-context.md codebase-onboarding skill
Product product-context.md product-strategist skill
RA/QM regulatory-context.md regulatory-affairs-head skill

Rules:

  • If context exists → read it, use it, only ask for gaps
  • If context doesn't exist → offer to create it (auto-draft from available info)
  • Never dump all questions at once — conversational, one section at a time
  • Push for verbatim language — exact customer/user phrases beat polished descriptions

Pattern 2: Practitioner Voice

Every skill opens with an expert persona and clear goal. Not a textbook — a senior practitioner coaching you.

Implementation:

You are an expert in [domain]. Your goal is [outcome].

Rules:

  • Write as someone who has done this 100 times
  • Use contractions, direct language
  • If something sounds like a Wikipedia article, rewrite it
  • Opinionated > neutral. State what works and what doesn't.
  • "Do X" beats "You might consider X"
  • Industry jargon is fine when talking to practitioners — explain when talking to founders

Anti-patterns:

  • ❌ "This skill provides comprehensive coverage of..."
  • ❌ "The following section outlines the various approaches to..."
  • ❌ "It is recommended that one should consider..."
  • ✅ "You are an expert in SaaS pricing. Your goal is to help design pricing that captures value."
  • ✅ "Lead with their world, not yours."
  • ✅ "If it sounds like marketing copy, rewrite it."

Pattern 3: Multi-Mode Workflows

Most skills have 2-3 natural entry points. Design for all of them.

Implementation:

## How This Skill Works

### Mode 1: Build from Scratch
When starting fresh — [describe the greenfield scenario].

### Mode 2: Optimize Existing  
When [artifact] exists but isn't performing. Analyze → identify gaps → recommend.

### Mode 3: [Situation-Specific]
When [edge case or specific scenario that needs a different approach].

Common mode pairs:

Skill Type Mode 1 Mode 2 Mode 3
CRO skills Audit a page Redesign flow A/B test specific element
Content skills Write new Rewrite/optimize Repurpose for channel
SEO skills Full audit Fix specific issue Competitive gap analysis
Strategy skills Create plan Review/critique plan Pivot existing plan
Analytics skills Set up tracking Debug tracking Analyze data

Rules:

  • Mode 2 (optimize) should ask for current performance data
  • If user has performance data → use it to inform recommendations
  • Each mode should be self-contained (don't assume they read the other modes)

Pattern 4: Related Skills Navigation

Every skill ends with a curated list of related skills. Not just links — when to use each and when NOT to.

Implementation:

## Related Skills

- **copywriting**: For landing page and web copy. NOT for email sequences or ad copy.
- **page-cro**: For optimizing any marketing page. NOT for signup flows (use signup-flow-cro).
- **email-sequence**: For lifecycle/nurture emails. NOT for cold outreach (use cold-email).

Rules:

  • Include 3-7 related skills (not all of them — curate)
  • Each entry: skill name + WHEN to use + WHEN NOT TO (disambiguation)
  • Cross-references must be bidirectional (A mentions B, B mentions A)
  • Include cross-domain references when relevant (e.g., marketing skill → business-growth skill)
  • Group by relationship type if >5: "Works with", "Instead of", "After this"

Pattern 5: Reference Separation

SKILL.md is the workflow. Reference docs are the knowledge base. Keep them separate.

Implementation:

skill-name/
├── SKILL.md              # ≤10KB — what to do, how to decide, when to act
├── references/
│   ├── frameworks.md      # Deep framework catalog
│   ├── benchmarks.md      # Industry data and benchmarks
│   ├── platform-specs.md  # Platform-specific details
│   └── examples.md        # Real-world examples
├── templates/
│   └── template.md        # User-fillable templates
└── scripts/
    └── tool.py            # Python automation

Rules:

  • SKILL.md ≤10KB — if it's longer, move content to references
  • SKILL.md links to references inline: See [references/frameworks.md](references/frameworks.md) for the full catalog.
  • References are loaded on demand — zero startup cost
  • Each reference doc is self-contained (can be read independently)
  • Templates are user-fillable files with clear placeholder markers

Pattern 6: Proactive Triggers

Skills surface issues without being asked when they detect patterns in context.

Implementation:

## Proactive Triggers

Surface these without being asked:

- **[Condition]**[What to flag and why]
- **[Condition]**[What to flag and why]

Rules:

  • 4-6 triggers per skill
  • Each trigger: specific condition + business consequence
  • Triggers should be things the user wouldn't think to ask about
  • Format: condition → flag → recommended action
  • Don't trigger on obvious things — trigger on hidden risks

Examples:

  • SEO: "Keyword cannibalization detected — two pages targeting the same term" → flag
  • Pricing: "Conversion rate >40% — likely underpriced" → flag
  • Content: "No content updated in 6+ months" → flag
  • CRO: "Form has >7 fields with no multi-step" → flag

Pattern 7: Output Artifacts

Map common requests to specific, concrete deliverables.

Implementation:

## Output Artifacts

| When you ask for... | You get... |
|---------------------|------------|
| "Help with pricing" | Pricing recommendation with tier structure, value metrics, and competitive positioning |
| "Audit my SEO" | SEO scorecard (0-100) with prioritized fixes and quick wins |

Rules:

  • 4-6 artifacts per skill
  • Each artifact has a specific format (scorecard, matrix, plan, audit, template)
  • Artifacts are actionable — not just analysis, but recommendations with next steps
  • Include what the output looks like (table? checklist? narrative?)

Pattern 8: Quality Loop

Skills self-verify before presenting findings.

Implementation:

## Communication

All output passes quality verification:
- Self-verify: source attribution, assumption audit, confidence scoring
- Peer-verify: cross-functional claims validated by the owning skill
- Output format: Bottom Line → What (with confidence) → Why → How to Act → Your Decision
- Results only. Every finding tagged: 🟢 verified, 🟡 medium, 🔴 assumed.

Rules:

  • Every finding tagged with confidence level
  • Assumptions explicitly marked as assumptions
  • "I don't know" > fake confidence
  • Cross-functional claims reference the relevant skill
  • High-stakes recommendations get extra scrutiny

Pattern 9: Communication Standard

Structured output format for all skill output.

Standard output:

BOTTOM LINE: [One sentence answer]

WHAT:
• [Finding 1] — 🟢/🟡/🔴
• [Finding 2] — 🟢/🟡/🔴

WHY THIS MATTERS: [Business impact]

HOW TO ACT:
1. [Action] → [Owner] → [Deadline]

YOUR DECISION (if needed):
Option A: [Description] — [Trade-off]
Option B: [Description] — [Trade-off]

Rules:

  • Bottom line first — always
  • Max 5 bullets per section
  • Actions have owners and deadlines
  • Decisions framed as options with trade-offs
  • No process narration ("First I analyzed...") — results only

Pattern 10: Python Tools

Stdlib-only automation that provides quantitative analysis.

Implementation:

#!/usr/bin/env python3
"""Tool description — what it does in one line."""

import json
import sys
from collections import Counter

def main():
    # Accept input from file arg or stdin
    # Process with stdlib only
    # Output JSON for programmatic use
    # Also print human-readable summary
    pass

if __name__ == "__main__":
    main()

Rules:

  • stdlib-only — zero external dependencies (no pip install)
  • CLI-first — run from command line with file args or stdin
  • JSON output — structured output for integration
  • Sample data embedded — runs with zero config for demo/testing
  • One tool, one job — focused, not Swiss Army knife
  • Scoring tools output 0-100 — consistent scale across all tools

Naming convention: snake_case_verb_noun.py (e.g., seo_checker.py, headline_scorer.py, churn_risk_scorer.py)


File Structure Standard

skill-name/
├── SKILL.md                    # ≤10KB — workflow, decisions, actions
├── references/                  # Deep knowledge (loaded on demand)
│   ├── [topic]-guide.md        # Comprehensive guide
│   ├── [topic]-benchmarks.md   # Industry data
│   └── [topic]-examples.md    # Real-world examples
├── templates/                   # User-fillable templates
│   └── [artifact]-template.md  # With placeholder markers
└── scripts/                     # Python automation
    └── [verb]_[noun].py        # Stdlib-only, CLI-first

Naming rules:

  • Skill folder: kebab-case
  • Python scripts: snake_case.py
  • Reference docs: kebab-case.md
  • Templates: kebab-case-template.md

Quality Checklist

Before a skill is considered done:

Structure

  • YAML frontmatter with name, description (trigger keywords), version
  • Practitioner voice — "You are an expert in X. Your goal is Y."
  • Context-first — checks domain context before asking questions
  • Multi-mode — at least 2 workflows (build/optimize)
  • SKILL.md ≤10KB — heavy content in references/

Content

  • Action-oriented — tells you what to do, not just what exists
  • Opinionated — states what works, not just options
  • Tables for structured comparisons
  • Checklists for processes
  • Examples for clarity

Integration

  • Related Skills section with WHEN/NOT disambiguation
  • Cross-references are bidirectional
  • Listed in domain CLAUDE.md
  • Listed in .codex/skills-index.json
  • Listed in .claude-plugin/marketplace.json
  • Listed in .gemini/skills-index.json (run ./scripts/gemini-install.sh)

Quality Standard

  • Proactive Triggers (4-6 per skill)
  • Output Artifacts table (4-6 per skill)
  • Communication standard reference
  • Confidence tagging on findings

Automation (if applicable)

  • Python tool(s) — stdlib-only, CLI-first, JSON output
  • Sample data embedded — runs with zero config
  • Scoring uses 0-100 scale

Domain Context Files

Domain File Sections
C-Suite company-context.md Stage, team, burn rate, competitive landscape, strategic priorities
Marketing marketing-context.md Brand voice, style guide, target keywords, internal links map, competitor analysis, audience personas, writing examples, customer language
Engineering project-context.md Tech stack, architecture, conventions, CI/CD, testing strategy
Product product-context.md Roadmap, personas, metrics, feature priorities, user research
RA/QM regulatory-context.md Device classification, applicable standards, audit schedule, SOUP list

Each domain's context skill creates this file via guided interview + auto-draft from available information.


This standard applies to all new skills and skill upgrades across the entire repository. Version: 1.0.0 | Created: 2026-03-06