| name | tokenmaxxing |
| description | Native token efficiency mode that compresses output, minimizes verbosity, and enables caveman-style communication. Use when the user wants to save tokens, reduce output length, work in minimal-verbosity mode, or mentions token budgets or caveman mode. |
| triggers | ["tokenmaxxing"] |
| scope | harness |
| invocation | workflow |
TOKENMAXXING MODE
Auto-Activation (full bake — 2026-04-25, verified 2026-05-14 PATCH 039)
token-session-init.js sets tokenmaxxing: true (L56) AND injects the full ## Rules block from this SKILL.md into Claude's startup context (L127) on every SessionStart. No manual command needed. Rules are live from turn 1.
/tokenmaxxing is preserved only as a re-activate alias after tokenmaxxing off mid-session. Subsequent docs should describe tokenmaxxing as "native default" rather than as a triggerable command.
Activation Steps (re-activate only — run when skill fires after tokenmaxxing off)
-
Run this Bash command to set the session flag:
node -e "const ss=require(process.cwd()+'/.claude/hooks/session-state.js'); ss.update(s=>{s.tokenmaxxing=true;});"
-
Output exactly one line: ⚡ TOKENMAXXING ACTIVE
Rules — enforce for the rest of this session
Ultra-Caveman
- Zero preamble. Zero trailing summaries. No "let me", no "I'll", no "here's".
- Single-sentence updates. Headers only for multi-section outputs.
- Code stays deep. Docs stay thorough. Speech stays minimal.
Auto-Compact
- When context hits tier 2 (60%+), run /compact immediately — do not suggest, do not wait.
- Use the pre-built hint injected by pre-tool-use.js context.
- Never compact without preserving: branch, files touched, last user correction, next step.
Tool Routing (native-first)
- Self-select model and fan-out per task — Opus 4.8 reasons the fit. No lane ladder, no external/local-model offload; those mechanisms are retired.
- Known/trivial reads →
Read/Grep/Bash directly. Never spawn an Agent to read a known file.
- Reserve
Agent/Workflow for genuine fan-out, cross-file reasoning, or write/tool-loops — sized to the task, no fixed cap.
Background Tasks (ctrl+b / [bg] prefix)
- If user input starts with
[bg] or /bg: spawn as Agent({ run_in_background: true }) immediately.
- Confirm in one line:
→ BG: <3-word description>
- Return to main thread. Do not narrate the spawned task further.
Deactivation
Type tokenmaxxing off to disable:
node -e "const ss=require(process.cwd()+'/.claude/hooks/session-state.js'); ss.update(s=>{s.tokenmaxxing=false;});"
Output: TOKENMAXXING OFF
Token Budget Policy
- Master enforcement: every active session enforces 5k–15k soft / 40k hard transcript budget.
- No command outputs > 60–80 lines.
- Dirty repos: no broad
git status or find .; scoped and marker-only only.
- Reports: pass = one label line; failure = failure-only verbose block.
- Runtime script flags planned separately; docs-only patches committed first.
Hard Token Rules — Research
- No routine subagents for package or web research. Direct shell only.
- No rendered GitHub WebFetch. Use raw.githubusercontent.com with line cap.
- Evidence pack max: 80 lines.
- Final report max: 120 lines unless explicitly blocked.
- Split before broad crawl. Full crawl requires explicit owner approval.
- DeepSeek reinforcement: compact evidence only. No raw dumps.
Session Notes
2026-06-13
- session: 116m | peak ctx: 0% | compacts: 0
- tools: Bash×947, Read×345, Edit×171, StructuredOutput×82, Write×63, TodoWrite×25, ToolSearch×8, Workflow×6, Agent×3, ScheduleWakeup×2, TaskStop×1, PushNotification×1, AskUserQuestion×1
- corrections: rick i have a fun little task for you. I will be giving you the task of going through trending repos on github, scanning them, compare yuri to those, see what we can adopt and rebuild better in yuri u
- errors: none
2026-06-10
- session: 47m | peak ctx: 64% | compacts: 1
- tools: Bash×278, Read×74, Write×15, Edit×11, WebFetch×2, Agent×1, TodoWrite×1
- corrections: alright insane work rick, I need all the fixes you propose to get the entire mathematical, scientific, physics, numerology, energy gate and more to work as one precisely engineered clockwork wherever | insane work rick. really absolutely insane. next up on the list (im looking at my yuri-chip-die for reference) is the entire 'memory/cognition/retrieval' which I need to be covered as a handover packa
- errors: none
2026-06-09
- session: 106m | peak ctx: 55% | compacts: 0
- tools: Bash×114, Read×104, Edit×40, WebFetch×9, Agent×6, Write×6, TodoWrite×4, ToolSearch×2
- corrections: none
- errors: none
2026-05-14
- session: 78m | peak ctx: 0% | compacts: 0
- tools: Bash×94, Read×20, Write×18, Edit×18, TodoWrite×6, ToolSearch×2, mcp×1, ExitPlanMode×1, Skill×1
- corrections: none
- errors: none
2026-05-02
- session: 4m | peak ctx: 14% | compacts: 0
- tools: Bash×16, Read×4, Edit×4, Skill×1
- corrections: none
- errors: none
2026-04-27
- session: 3m | peak ctx: 50% | compacts: 0
- tools: Bash×8, Read×5
- corrections: none
- errors: none
2026-04-26
- session: 7m | peak ctx: 0% | compacts: 0
- tools: Bash×15, Read×9, Write×4, Agent×1, ToolSearch×1, ExitPlanMode×1, Edit×1
- corrections: none
- errors: none
2026-04-25
- session: 4m | peak ctx: 16% | compacts: 0
- tools: Bash×4, Edit×4, ToolSearch×1, ExitPlanMode×1, Read×1
- corrections: none
- errors: none