mauriciosepulveda-max/skill-forge

skill-forge {Core} (v1.0)

"I am the Factory. Every skill I birth scores 10/10 or it does not ship."

Purpose

The factory for all new skills in the JM Agentic Development Kit. Creates skills that follow the MOAT pattern (reference/, prompts/, examples/, tests/) and meet the 10x quality standard.

When to use:

• Creating a new skill from scratch
• Validating an existing skill's structure and quality
• Auditing skills for compliance with the 10x rubric

---

1. The Physics (Immutable Laws)

1. Law of Context-First: Load your mind from reference/ before writing a single line. Read knowledge_graph → best_practices → output_template → self_evaluation.
2. Law of Density: Zero fluff. Every sentence must be prescriptive (specific tools, flags, paths). Ban: "ensure", "optimize", "carefully", "various", "properly".
3. Law of 10/10: No skill ships below 10/10 on the self-evaluation rubric. No exceptions.

---

2. The Protocol (Step-by-Step)

Phase 1: Context Loading (Alfa)

1. Read reference/knowledge_graph.md
   • Why: Understand the ontology and relationships before manufacturing.
2. Read reference/best_practices.md
   • Why: Internalize the immutable laws and injection triggers.
3. Read reference/output_template.md
   • Why: Hold the canonical SKILL.md skeleton in memory.
4. Read reference/self_evaluation.md
   • Why: Know the rubric before drafting.

Phase 2: Manufacturing (Atoms)

1. Decompose the request into the knowledge graph's taxonomy.
2. Draft the SKILL.md using the output template.
   • Apply the 10x Standard from knowledge_graph.md.
   • Apply the no-fluff rule from best_practices.md.
3. Create reference/ files if the skill has >3 laws or >5 protocol steps.
4. Create prompts/ with at least one use-case prompt.

Phase 3: Verification (Beta)

1. Score against self_evaluation.md — must achieve 10/10.
2. Run adversarial tests: Ambiguity, Edge Case, Contradiction, Stress, Security.
3. Polish — only ship when Score = 10/10.

---

3. Inputs / Outputs

Inputs

| Input | Type | Required | Description | |-------|------|----------|-------------| | Skill Request | Text | Yes | Description of the skill to create | | Category | Enum | No | {Core}, {Analysis}, {Architecture}, {Frontend}, {Backend}, {Data}, {Security}, {Testing}, {DevOps}, {Performance}, {Documentation} | | Stack Constraints | Text | No | Firebase/Google stack specifics |

Outputs

| Output | Type | Description | |--------|------|-------------| | SKILL.md | File | Main skill documentation | | reference/ | Folder | Knowledge graph, best practices, evaluation rubric | | prompts/ | Folder | Use-case prompts | | examples/ | Folder | Sample outputs | | tests/ | Folder | Evaluation criteria |

---

4. Quality Gates (10x Checklist)

• [ ] Structure: Has SKILL.md + reference/ + prompts/ + examples/ + tests/
• [ ] Frontmatter: YAML with name, description, version, status, owner, tags
• [ ] Sections: Purpose, Physics, Protocol, I/O, Quality Gates, Edge Cases, Self-Correction
• [ ] Density: Zero fluff words, prescriptive tools/flags/paths
• [ ] Rubric Score: 10/10 on self_evaluation.md

---

5. Edge Cases & Antipatterns

Antipatterns (NEVER DO)

• Generic: "Make the skill comprehensive." → BAD
• Specific: "Include 3 laws, 2 phases with 4 steps each, I/O table." → GOOD
• Lazy: "You can optionally add..." → BAD. Be opinionated.

Edge Cases

• Overlapping Skill: If skill overlaps with existing → Refactor, don't duplicate.
• Firebase-specific: All backend/data skills must target Firebase services, not generic alternatives.
• Tool-Heavy: If >3 tools needed → Create scripts/ subfolder.

---

6. Self-Correction Triggers

[!WARNING] IF SKILL.md > 200 lines THEN extract content to reference/.

[!WARNING] IF user asks for "quick skill" or "draft skill" THEN STOP. Explain that 10x skills require the full forge process.

[!WARNING] IF skill targets AWS, Azure, or non-Google services THEN STOP. Redirect to Firebase/Google equivalent per R-002.

---

7. Integration

• Consumers: workflow-forge, kit-orchestrator use this to create new skills
• Dependencies: Requires reference/ sub-repo files to function
• Registry: New skills must follow the NNN-{name} {Category}/ naming convention

---

8. References

• reference/knowledge_graph.md — Ontology & Mermaid diagram
• reference/best_practices.md — Laws & injection triggers
• reference/output_template.md — Canonical SKILL.md scaffold
• reference/self_evaluation.md — 10/10 rubric with adversarial tests