# Skill Architect Protocol

> Loaded by the skill-architect agent to design or revise Superpipelines-compatible skills.

- Skill: `gustavo-meilus/skill-architect-protocol` (Agent Skill)
- Install (CLI): `npx skillmds@latest add gustavo-meilus/skill-architect-protocol`
- Raw SKILL.md: https://api.skillmd.com/api/skills/gustavo-meilus/skill-architect-protocol/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: gustavo-meilus (https://skillmd.com/u/gustavo-meilus)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/gustavo-meilus/skill-architect-protocol

---


# Skill Architect — Operational Protocol

<overview>
The Skill Architect treats every skill as a precision instrument, scoped to a single capability and optimized for description-based routing. It enforces the separation of high-level workflow from deep reference content to maintain lean system contexts and ensure reliable execution.
</overview>

<glossary>
  <term name="Routing Contract">The skill's `description` field, used by the model to decide when to invoke the body.</term>
  <term name="Progressive Disclosure">Structuring skills into a high-level `SKILL.md` and detailed `references/*.md` to manage context density.</term>
  <term name="Rationalization Resistance">Mechanisms (Red Flags, STOP sections) that prevent the model from ignoring constraints.</term>
</glossary>

## Operating Modes

<operating_modes>
| Mode | Trigger | Outputs |
| :--- | :--- | :--- |
| **ARCHITECT** | Prompt: "Design a skill for..." | Skill directory, `SKILL.md`, references, and Architect's Brief. |
| **QUICK-BUILD** | Prompt: "Quick skill for X." | Minimal skill directory and `SKILL.md`. |
| **DIAGNOSE** | Prompt: "Review this skill." | Architecture score and top-3 remediation steps. |
| **EXTRACT** | Prompt: "Capture this workflow." | New skill extracted from conversation history. |
| **ITERATE** | Prompt: "Improve this skill." | Edited artifacts with a change summary. |
</operating_modes>

## Protocol

<protocol>
### 1. DECONSTRUCT
- Run the 4D Method internally; gate execution if ≥3 critical slots (capability, scope, triggers) are missing.
- Glob existing skills to prevent name collisions.
- Identify the single core capability of the requested skill.

### 2. DESIGN
- **Invocation Model**: Set `disable-model-invocation` and `user-invocable` based on use case (Reference, User-only, or Workflow).
- **Topology**: Choose between a monolithic `SKILL.md` (<500 lines) or a layered structure using `references/` for deep content.
- **Description**: Write the description FIRST as it defines the routing contract.
- **Resistance**: Design Red Flags and a Rationalization Table for discipline-enforcing skills.

### 3. DEVELOP
- Create the skill directory and `SKILL.md` with appropriate frontmatter.
- Follow the `references/skill-architecture.md` template for body structure.
- **References**: Ensure any reference file >100 lines includes a Table of Contents.

### 4. DELIVER
- Provide the Architect's Brief detailing the invocation model and key decisions.
- Include 3–5 test prompts to verify triggering accuracy.
- Offer a deployment checklist (RED baseline / GREEN with skill).
</protocol>

<invariants>
- **Description Rules**: Triggering conditions only; third-person voice; ≤1536 characters (combined with `when_to_use` if present); no workflow summaries.
- **Content Rules**: Bodies must not exceed 500 lines; use concrete examples over abstract rules.
- **Safety**: Skills must never contain security-compromising or malicious content.
- **References**: Maintain one level of nesting (no subdirectories within `references/`).
</invariants>

## Reference Files

- `${CLAUDE_PLUGIN_ROOT}/skills/skill-architect-references/references/skill-architecture.md`
- `${CLAUDE_PLUGIN_ROOT}/skills/skill-architect-references/references/description-engineering.md`
- `${CLAUDE_PLUGIN_ROOT}/skills/skill-architect-references/references/claude-code-skill-spec.md`
- `${CLAUDE_PLUGIN_ROOT}/skills/skill-architect-references/references/anti-patterns.md`