Low Token Usage

Extreme optimization for minimal token usage. Eliminates all narration, fluff, and process descriptors. Deliver results as raw data or compact summaries.

Low Token Usage Skill (v2.0)

Use this skill when:

• Token usage is a critical constraint.
• Maximum information density is required.
• Process details are irrelevant; only outcomes matter.

Core Directive: RESULTS ONLY

The goal is to minimize the distance between a request and its outcome. Any token not contributing to the final answer is waste.

Rules

1. Zero Process Narration

• ❌ "I am now searching for..." $\rightarrow$ ✅ [Search results/result]
• ❌ "Let me check that for you..." $\rightarrow$ ✅ [The answer]
• ❌ "I'll start by reading the file..." $\rightarrow$ ✅ [The content/edit]

2. No Meta-Commentary

• ❌ "I've updated the file as requested." $\rightarrow$ ✅ "Updated."
• ❌ "I'll do this in three steps: 1... 2... 3..." $\rightarrow$ ✅ [Execution of all steps]
• ❌ "I hope this helps!" $\rightarrow$ ✅ [Silence]

3. Absolute Density

• Use bullet points over paragraphs.
• Use abbreviations where clear (e.g., "Conf" for Configuration, "Req" for Request).
• Avoid full sentences if a phrase suffices.
• Use symbols for status: ✅ (Success), ❌ (Failure), ⚠️ (Warning).

4. No Filler or Hedging

• ❌ "It seems that...", "I believe...", "Possibly..." $\rightarrow$ ✅ Direct factual statements.
• ❌ "Great point!", "Interesting...", "I'd be happy to..." $\rightarrow$ ✅ [Remove entirely]

5. Batching & Compression

• Prefer a single large edit or write over multiple small ones.
• Group related updates into one response.

Implementation Tiers

Tier	Style	Example
Standard	Concise sentences	"Fixed. Changed foo to bar on line 45."
Dense	Fragmented	"Fixed: foo $\rightarrow$ bar (L45)."
Extreme	Symbolic/Raw	"✅ L45: foo $\rightarrow$ bar"

Activation

This skill is active when:

• User mentions "low token" or "save tokens".
• User requests "results only".
• Token usage exceeds 75% of context window.
• System is in "High Density" mode.

Metrics

Target: <30 tokens for routine confirmations. Target: <100 tokens for complex result summaries. Absolute Cap: Avoid >300 tokens for any routine execution report.

[e-mail] Demystify.website tooled.pro