Skill Factory

🔒 Prerequisites (Mandatory)

This skill acts as the architect for the Hub's capabilities. Before execution: 0. Mode Check: Verify .hub-mode and apply token-distiller guidelines.

1. Context Check: Rehydrate state by reading .specs/project/STATE.md, .specs/project/MEMORY.md, and .specs/project/LEARNINGS.md.
2. Knowledge Check: Follow the Knowledge Verification Chain (see below).

Skill Factory: Capability Architect (v2.3.0)

"Structure is the bridge to autonomy. Precision in design, rigor in execution."

🧩 Delegation Matrix

The Skill Factory delegates specific lifecycle tasks to its sub-agents:

🏗️ Skill Authoring Patterns (The Core)

Group 1: Discovery & Selection

1. Activation Metadata: Precise descriptions that trigger the agent correctly.
2. Exclusion Clause: Explicit "Do NOT use for..." lines to prevent hijacking.

Group 2: Context Economy

3. Context Budget: Every sentence must justify its token cost.
4. Progressive Disclosure: Move details to sub-files (REFERENCE.md, FORMS.md).

Group 3: Instruction Calibration

5. Control Tuning: Match instruction freedom to task fragility.
6. Explain-the-Why: State rules + rationale for better generalization.
7. Template Scaffold: Provide explicit placeholders for reports/logs.
8. In-Skill Examples: Embed 2-3 Few-shot pairs for style calibration.
9. Known Gotchas: Document failure modes to prevent hallucinations.

Group 4: Workflow Control

10. Execution Checklist: Copyable checklists to track progress.
11. Self-Correcting Loop: "Produce-validate-fix" loops for quality.
12. Plan-Validate-Execute: Verified plan before performing side effects.

Group 5: Executable Logic

13. Utility Bundle: Purpose-built scripts for deterministic logic.
14. Autonomy Calibration: Declare allowed-tools to minimize friction.

🔄 4-Phase Workflow

1. DISCOVERY

• Goal: Understand the functional domain and intent.
• Action: Use skill-factory-bootstrap to perform an Intent Audit.
• Output: Defined triggers and core task scope.

2. SPECIFY

• Goal: Design the skill's soul and instructions.
• Action: Apply Control Tuning and Explain-the-Why. Define the sub-file structure.
• Output: Initial draft of SKILL.md and required sub-files.

3. IMPLEMENT

• Goal: Calibrate and finalize instructions.
• Action: Add Few-Shot Examples and Known Gotchas.
• Output: Verified skill directory.

4. VERIFY

• Goal: Audit against SDD and Slop-Free standards.
• Action: Use skill-factory-validator to produce an audit score.
• Output: Final audit-report.md and registration.

🛠️ Operational Protocols

1. The "Structural Purity" Mandate

Every skill created MUST be self-documenting and logically isolated:

• Explicit Logic: No hidden dependencies or "magic" CLI commands.
• Zero-Fragment Memory: Skills MUST NOT store their own STATE.md, MEMORY.md, LEARNINGS.md, or tasks.md. All operational memory MUST be aggregated in .specs/project/ or .specs/features/.
• Atomic Intent: Every task and spec must have a clear ID.

2. Knowledge Verification Chain

1. Core SDD Skill: Ultimate truth for workflow patterns.
2. Existing Skills: High-performance references (e.g., sdd, git-workflow).
3. Internal Governance: .specs/codebase/GLOBAL_MANDATES.md.

3. Safety Valve

If a skill design exceeds 5 sub-skills or complex external dependencies, STOP and re-evaluate for orchestration.

4. Observable Governance (v2.3.0)

Every skill created MUST support the automated auditing protocol.

• Metadata: Append <!-- @sdd-state --> to all .md files.
• Task Evidence: Tasks for skill development MUST reside in .specs/features/ and include an evidence-based table.
• Discovery: Ensure the skill folder contains a valid SKILL.md for bin/check-health.py discovery.

🚫 Prohibited

• NEVER modify system directories (e.g., .agents/, .gemini/) unless explicitly authorized.
• NEVER create local STATE.md, MEMORY.md, or LEARNINGS.md within a skill; use .specs/project/ instead.
• NEVER approve a skill that doesn't follow the 4-Phase SDD workflow.
• NEVER use placeholders like "todo" or "..." in instructions.

Law of the Factory: A skill without structure is just a prompt. Structure is the bridge to autonomy.

version: "2.3.0"
feature_id: "HUB-ALIGNMENT"
phase: "VERIFY"
status: "COMPLETED"
last_update: "2026-05-06T09:48:00Z"
evidence_checksum: "8e52f6a"