skill-architect-protocol

name	skill-architect-protocol
description	Loaded by the skill-architect agent to design or revise Superpipelines-compatible skills.
disable-model-invocation	true
user-invocable	false

Skill Architect — Operational Protocol

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. The skill's `description` field, used by the model to decide when to invoke the body. Structuring skills into a high-level `SKILL.md` and detailed `references/*.md` to manage context density. Mechanisms (Red Flags, STOP sections) that prevent the model from ignoring constraints.

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

### 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).

- **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/`).

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