What problem does it solve? Writing steering files for AI coding assistants (skills, subagents, instructions, prompts) requires version-specific frontmatter fields, description shapes, and discovery budgets that differ across Claude Code, Copilot, Codex, and APM — getting them wrong from memory produces files that never activate or fail to load. ## Core Features & Use Cases - Customization type routing: A decision table that resolves whether a new rule belongs in a skill, agent, instruction, prompt, hook, MCP server, or plugin before any file is written. - Description craft guidance: Evidence-based rules for writing discovery descriptions (directive phrasing, negative constraints, naming) plus the four context budgets that govern skill visibility across harnesses. - Cross-harness frontmatter parity: Field mappings for dual-deployed files (applyTo vs paths, tools vs disallowedTools, model formats) and a shared provenance convention with license-fidelity rules. - Use Case: When asked to create a new subagent for code review, load this skill first to pick the correct file type, write a compliant frontmatter, and route to the depth references for agent tools, handoffs, and orchestration. ## Quick Start Use the meta-steering skill to author a new SKILL.md for a PDF-processing workflow with correct frontmatter and description.
