add-token-efficiencylisted
Install: claude install-skill brabos-ai/code-addiction
# Token Efficiency
<!-- uses:
- skill: add-claude-md-style
- skill: add-doc-schemas
- mention: add-id-convention
-->
## Overview
Every token counts. Resources must be compressed without losing clarity. This skill defines mandatory compression patterns for all ADD outputs (commands, skills, scripts, docs).
For doc-specific concerns (schemas, depth floors, output-length doctrine, validation gate), see `{{skill:add-doc-schemas/SKILL.md}}`. Output length is governed there — this skill covers compression *patterns* only.
## When NOT to use
- Output-length decisions (word/char caps, depth floors) → use `{{skill:add-doc-schemas/SKILL.md}}`
- Prose voice — tone, figurative language, what a sentence may say → use `{{skill:add-doc-schemas/SKILL.md}}`
- CLAUDE.md style and content rules → use `{{skill:add-claude-md-style/SKILL.md}}`
- ID format / branch naming → use `{{skill:add-id-convention/SKILL.md}}`
---
## Mandatory Patterns
### 1. JSON Specs (Always Minified)
```markdown
# BAD - 156 chars
```json
{
"type": "command",
"trigger": "slash",
"structure": "phases"
}
```
# GOOD - 48 chars
{"type":"cmd","trigger":"slash","structure":"phases"}
```
### 2. JSON = DATA. Markdown = INSTRUCTIONS.
JSON minified is exclusively for **data the agent looks up**: configs, specs, tech stack, path mappings, key-value pairs. Everything that orients behavior — rules, instructions, guidance, constraints, conventions — uses **markdown** (lists, tables, prose).
**Minified JSON — structur