context-optimization-auditlisted

# Claude Code Context Optimization Audit Audit all loaded context sources and propose what to cut, merge, condense, disable, or move. **Do not change anything first.** Produce a plan, wait for approval, then apply only approved changes. ## Goal Reduce token usage while preserving behavior. Lower duplication, clearer triggers, fewer competing rules, better global/project/task separation. ## Core principle More instructions is not better. Keep always-loaded context short. Move rarely-used detail into narrowly triggered skills, commands, or reference docs. ## What to inspect Search all of these (do not assume paths are complete — use `find`, `ls`, `rg`): - `CLAUDE.md`, `AGENTS.md` - `.claude/`, `.claude/skills/`, `.claude/agents/`, `.claude/commands/`, `.claude/settings.json` - `.mcp.json`, `package.json` (Claude/MCP-related scripts) - Any `agents.json`, `skills.json`, `opencode.json`, or migrated legacy config files - `~/.claude/` (user-level config, if accessible) - Project docs repeatedly copied into prompts or agent instructions ## Audit process ### 1. Inventory For each context source, capture: type · path · size (lines + `chars/4` token estimate) · purpose · trigger · global vs project · recently used or obsolete · overlap · risk · initial recommendation. ### 2. Find duplication and overlap Check for: - Same workflow in `CLAUDE.md`, skills, and commands simultaneously - Old JSON agent definitions alongside current Markdown versions - Broad skills with imprecis