← Back to wkappbot-workflow
wkappbot-workflow
★★★
Claude Token Optimization — Exploration, Context & Search
80-90% of tokens are spent on file exploration, not writing code. Techniques: (1) skill-search-first prevents redundant Glob/Grep. (2) .claudeignore + lean CLAUDE.md + session mgmt saves 30-40%. (3) PreToolUse hook enforces search discipline globally. (4) Codebase pre-indexing saves 90%+ of exploration tokens. (5) 3-agent team eliminates scope creep.
user/developer
tokenoptimizationclaudeignoreclaude.mdsessioncompactpre-index3-agent
Steps
- [SKILL-SEARCH-FIRST] MANDATORY before any Gl...
- [PRETOOLUSE-HOOK] Enforce skill-search disci...
- [CLI-SKILL-HINT] PrintSkillSearchHint() embe...
- [CLAUDEIGNORE] Create .claudeignore in proje...
- [CLAUDEIGNORE-HQ] WKAppBot-specific .claudei...
- [CLAUDE-MD-200] CLAUDE.md hard cap = 200 lin...
- [SESSION] Use /compact when conversation gro...
- [MCP] Only connect MCP servers you actively...
- [QMD-MODES] Three search modes: (1) qmd sear...
- [QMD-TIPS] Search without --collection searc...
- [3-AGENT] Split work into 3 roles: (1) Archi...
- [3-AGENT-WHY] Without role separation, Claud...
- [CLAUDE-MD-RULES] 5 rules to embed: (1) Neve...
- [PRIORITY] Apply in order: skill-search-firs...
- [IMPACT] Combined: same token budget handles...