Standardize and Optimize AI Skills

This skill acts as a dispatcher for skill-creation excellence in the OpenCode environment, routing to the right template and standard based on a decision tree rather than providing one-size-fits-all guidance. For creating a new skill, it routes by complexity: Tier 1 for simple single-file skills under 200 lines, Tier 2 for complex multi-concept skills (200-1000 lines), and Tier 3 for massive platform skills (10+ products, e.g. AWS or Convex). For improving an existing skill, it routes by symptom: "too long" points to modularizing into Tier 3 structure, "AI ignores rules" points to the Anti-Rationalization technique, and "users can't find it" points to CSO - search optimization for LLMs, essentially "SEO for LLMs," writing descriptions that actually trigger. For verifying compliance, it points to metadata/naming Standards and a Testing Guide.

It provides four skill templates (Technique/how-to, Reference/docs, Discipline/rules, Pattern/design-patterns) and a concrete five-step process: identify the goal via the decision tree, select the matching template, apply CSO to optimize the description for discovery, add anti-rationalization content for discipline-type skills, and test using a RED-GREEN-REFACTOR cycle. It documents common mistakes with fixes - a description that summarizes the workflow instead of using "Use when..." trigger language, missing metadata.triggers (should have 3+ keywords), a generic name like "helper" instead of a gerund form like "creating-skills," and a long monolithic SKILL.md that should be split into references/. A pre-deploy checklist covers exact name-to-directory matching, all-caps SKILL.md filename, a "Use when..." description opener, 3+ trigger keywords, staying under 500 total lines, avoiding @ force-loading in cross-references, and testing with real scenarios.

Use this skill when creating a new skill from scratch, improving an existing skill that agents are ignoring, debugging why a skill isn't being triggered, or standardizing skill quality across a team. A companion /write-skill command offers guided skill creation, and opencode-expert is the related skill for OpenCode environment configuration itself.

It is specifically for authoring skills within the OpenCode skill-file convention (YAML frontmatter, SKILL.md, tiered architecture) - it is not a general writing-improvement or prose-editing skill despite the "Writing Skills" name.

Input is a skill-authoring task - creating a new skill, fixing one that's ignored or hard to discover, or verifying compliance before deployment. Output is the matching template selected via the decision tree, a description optimized with CSO principles, anti-rationalization content where needed, and a completed pre-deploy checklist confirming naming, metadata, length, and testing requirements are met.

---
name: my-technique
description: Use when [specific symptom occurs].
metadata:
  category: technique
  triggers: error-text, symptom, tool-name
---

Built around the OpenCode skill-file system (~/.config/opencode/skills/, SKILL.md, YAML frontmatter with metadata.triggers), with a /write-skill guided command and cross-referenced references/ documentation for CSO, Standards, Anti-Rationalization, and Testing.

Skill authors and teams building or maintaining a library of AI agent skills who need consistent architecture, discoverable descriptions, and rules agents actually follow rather than ignore.