micro-skill-creator

L1 Improvement

• Restructured the micro-skill playbook using the Skill Forge required sections and explicit completion gates.
• Added prompt-architect constraint extraction, confidence ceilings, and examples to avoid over-broad micro-skills.

---

LIBRARY-FIRST PROTOCOL (MANDATORY)

Before writing ANY code, you MUST check:

Step 1: Library Catalog

• Location: .claude/library/catalog.json
• If match >70%: REUSE or ADAPT

Step 2: Patterns Guide

• Location: .claude/docs/inventories/LIBRARY-PATTERNS-GUIDE.md
• If pattern exists: FOLLOW documented approach

Step 3: Existing Projects

• Location: D:\Projects\*
• If found: EXTRACT and adapt

Decision Matrix

Match	Action
Library >90%	REUSE directly
Library 70-90%	ADAPT minimally
Pattern exists	FOLLOW pattern
In project	EXTRACT
No match	BUILD (add to library after)

---

STANDARD OPERATING PROCEDURE

Purpose

Produce narrowly-scoped skills that do one job exceptionally well, with clear inputs/outputs, examples, and validation evidence for easy composition.

Trigger Conditions

• Positive: need a reusable micro-skill, want to refactor a large skill into smaller parts, or require fast iterations on a narrow task.
• Negative/reroute: full agent creation (agent-creator), broad prompt tuning (prompt-architect), or full skill scaffolding (skill-builder/skill-forge).

Guardrails

• Keep the scope single-purpose; avoid feature creep.
• Provide English outputs with explicit confidence ceilings and deterministic formats.
• Include at least one example and one edge-case check.
• Document how this micro-skill composes with others and where it should NOT be used.

Execution Phases

1. Intake: Capture the atomic task, constraints, success criteria, and consumers; classify hard/soft/inferred constraints.
2. Design: Define inputs/outputs, preconditions, and failure behaviors; choose minimal dependencies.
3. Author: Write the SKILL.md content, examples, and tests.
4. Validate: Run happy/edge tests; record evidence and ceilings.
5. Publish: Provide usage notes and composition guidance; register with meta-tools/skill-forge index if needed.

Pattern Recognition

• Data formatting/conversion → deterministic IO, schema validation.
• Guard/validation micro-skills → explicit error codes and refusal policy.
• Retrieval micro-skills → caching hints and rate-limit safety.

Advanced Techniques

• Use table-driven prompts to keep behavior predictable.
• Apply few-shot examples demonstrating both success and refusal paths.
• Add lightweight observability (logs/notes) for downstream debugging.

Common Anti-Patterns

• Multi-purpose scope disguised as micro-skill.
• Missing examples or unclear error handling.
• Over-reliance on assumptions instead of explicit constraints.

Practical Guidelines

• Keep the SKILL.md concise; link to references/resources for depth.
• Mark optional parameters and defaults clearly.
• Encourage idempotent behaviors where possible.

Cross-Skill Coordination

• Upstream: prompt-architect for clarity; cognitive-lensing for alternative framings.
• Downstream: meta-tools for composition; agent-creator for embedding into agents; recursive-improvement for tuning.

MCP Requirements

• Optional memory MCP to store usage notes; tag WHO=micro-skill-creator-{session}, WHY=skill-execution.

Input/Output Contracts

inputs:
  task: string  # required atomic task
  constraints: list[string]  # optional constraints
  io_format: string  # optional preferred format
outputs:
  skill_spec: file  # SKILL.md content for the micro-skill
  examples: list[file]  # example invocations
  tests: list[file]  # validation checks

Recursive Improvement

• Run recursive-improvement after initial use to capture failure cases and tighten scope; stop when delta < 2% or risks noted.

Examples

• Create a JSON schema validator micro-skill with refusal path for invalid input.
• Build a markdown-to-summary micro-skill with length and tone controls.

Troubleshooting

• Skill too broad → split into smaller micro-skills.
• Unclear errors → add structured error outputs and refusals.
• Flaky results → add few-shot anchors and deterministic formatting.

Completion Verification

• Scope single-purpose and documented.
• Examples and tests included with evidence.
• Confidence ceiling stated; inputs/outputs explicit.
• Composition guidance provided.

Confidence: 0.70 (ceiling: inference 0.70) - Micro-skill SOP rewritten with Skill Forge scaffolding and prompt-architect ceilings.