[Proposal] 4 additions: Skill vs MCP matrix, Conditional Workflows, Output Patterns, Distribution

[MainFrame100]

Hey! Great work on skill-conductor — the TDD cycle and scoring system are solid.

We've been building and maintaining 20+ custom skills for Claude Code over the past 5 months. Along the way we've hit pain points that your conductor doesn't cover yet. Below are 4 concrete proposals with ready-to-use markdown.

---

Proposal 1: Phase 0 — Skill vs MCP Decision Matrix

Problem: People start building a skill when they actually need an MCP server (or vice versa). This wastes the entire CREATE cycle.

Where: SKILL.md → CREATE mode, insert before Phase 1 as "Phase 0: Approach Selection"

Ready-to-insert content:

Phase 0: Approach Selection

Before creating a skill, determine if a skill is the right approach:

Approach	When to use	Example
MCP only	Simple tool access, ad-hoc queries, data exploration	"Query my database", "Read Slack messages"
Skill only	Repeatable workflows, domain expertise, consistent methodology	brand-guidelines, code-review-checklist
Skill + MCP	Complex workflows over external services, methodology atop raw tools	sentry-code-review, sprint-planner-with-linear

Decision questions:

1. Will this workflow be repeated more than 3 times? → Skill
2. Is it just connecting to an external service? → MCP
3. Does it need consistent methodology on top of service access? → Skill + MCP

Red flags you need MCP, not a skill:

• Skill body is 80% "call this API endpoint"
• No domain logic, just data plumbing
• Every run needs different parameters with no pattern

Red flags you need a skill, not just MCP:

• Users don't know what to do after connecting MCP
• Every conversation starts from scratch
• Results vary wildly between sessions

Success: Clear decision on approach BEFORE investing in skill creation.

---

Proposal 2: Conditional Workflows Pattern

Problem: references/patterns.md has 5 patterns but no branching/conditional logic pattern. Many skills need "if X → workflow A, else → workflow B" — this is one of the most common patterns in multi-mode skills (like conductor itself!).

Where: references/patterns.md → add as Pattern 6

Ready-to-insert content:

Pattern 6: Conditional Workflows

Best for: tasks with branching logic based on context or input type.

1. Determine the modification type:
   **Creating new content?** → Follow "Creation workflow" below
   **Editing existing content?** → Follow "Editing workflow" below

2. Creation workflow: [steps]
3. Editing workflow: [steps]

Key: "clear decision criteria at branch points, named workflow paths, no ambiguity about which path to follow."

Real example — a skill that handles both new docs and editing:

Step 1: Detect intent
- If user provides existing file path → EDIT mode
- If user describes desired output → CREATE mode
- If ambiguous → ask user

EDIT mode:
1. Read existing file
2. Identify what to change
3. Apply changes preserving structure

CREATE mode:
1. Gather requirements
2. Select template
3. Generate output

Anti-pattern: mixing branch logic with sequential steps. Keep branches named and separate.

---

Proposal 3: Output Patterns (new reference file)

Problem: No guidance on output consistency. Skills often exist specifically to produce consistent output — report formats, commit styles, document structures. Without this, agents improvise every time.

Where: New file references/output-patterns.md

Ready-to-insert content:

Template Pattern

Provide templates for output format. Match strictness to requirements.

Strict (API responses, data formats):

ALWAYS use this exact template:

# [Analysis Title]

## Executive summary
[One-paragraph overview of key findings]

## Key findings
- Finding 1 with supporting data
- Finding 2 with supporting data

## Recommendations
1. Specific actionable recommendation
2. Specific actionable recommendation

Flexible (when adaptation is useful):

Sensible default format — use your best judgment:

# [Analysis Title]

## Executive summary
[Overview]

## Key findings
[Adapt sections based on what you discover]

## Recommendations
[Tailor to the specific context]

Examples Pattern

For skills where output quality depends on seeing examples. Input/output pairs beat descriptions:

Generate commit messages following these examples:

Example 1:
Input: Added user authentication with JWT tokens
Output: feat(auth): implement JWT-based authentication

Example 2:
Input: Fixed bug where dates displayed incorrectly
Output: fix(reports): correct date formatting in timezone conversion

When to use which

Need	Pattern	Freedom level
Exact format required	Template (strict)	Low
General structure, flexible details	Template (flexible)	Medium
Style/tone consistency	Examples	Medium
No format requirements	None needed	High

---

Proposal 4: Distribution Section

Problem: PACKAGE mode ends with a .skill zip file. But what next? Users need to know distribution channels.

Where: SKILL.md → after Mode 5: PACKAGE, or as a subsection within it

Ready-to-insert content:

Distribution Channels

After packaging:

Channel	How	Best for
Local file	Share .skill file directly	Personal use, small teams
Skills API	/v1/skills endpoint, container.skills in Messages API	Programmatic deployment
Organization	Admin workspace deployment (auto-updates to all users)	Enterprise teams
Open repos	GitHub, awesome-claude-skills lists	Community sharing

The .skill format (zip) is an open standard — skills are portable across Claude Code, Claude.ai, and compatible platforms.

---

Summary

#	Proposal	Where	Effort
1	Skill vs MCP Matrix	SKILL.md, new Phase 0	~30 lines
2	Conditional Workflows	references/patterns.md	~25 lines
3	Output Patterns	new references/output-patterns.md	~60 lines
4	Distribution	SKILL.md, after PACKAGE	~15 lines

Happy to submit a PR with all of this if you're interested. All content is MIT-compatible.