From asi
Minimizes token costs in Claude Code by enforcing terse responses, parallel tool calls, targeted file reads, and subagent isolation for efficient sessions.
npx claudepluginhub plurigrid/asi --plugin asiThis skill uses the workspace's default tool permissions.
**Trit**: -1 (MINUS - validator/constrainer)
Optimizes Claude Code token usage with bash/grep/glob preferences over full file reads, Sonnet for dev/debugging, Opus for codebase learning, and minimal output strategies.
Reduces token waste by 40-60% with anti-sycophancy rules, tool-call budgets, one-pass coding, task profiles, and read-before-write enforcement. For verbose or expensive sessions.
Manages Claude Code token budgets with MCP-first navigation, lazy loading, subagent isolation, and strategic file reading. Use for long sessions, sluggish performance, or large codebases.
Share bugs, ideas, or general feedback.
Trit: -1 (MINUS - validator/constrainer) Purpose: Minimize Amp thread costs through token efficiency
| Pattern | Cost | Fix |
|---|---|---|
| Reading entire files | High | Use line ranges [1, 50] |
| Sequential tool calls | Medium | Parallelize independents |
| Explaining before doing | Medium | Just do it |
| Asking permission | Low-Medium | Act, don't ask |
| Repeating user's question | Low | Skip acknowledgment |
| Long error explanations | Medium | Terse: "Error: X. Fix: Y" |
| Multiple edit iterations | High | Plan first, single edit |
| Loading unused skills | Medium | Load on-demand |
# Bad: Read full 2000-line file
Read("/path/to/big.py")
# Good: Read relevant section
Read("/path/to/big.py", [100, 150])
# Better: Grep first, then targeted read
Grep("def target_function", path="/path/to/big.py")
Read("/path/to/big.py", [142, 165])
# Bad: Sequential
Read(file1) → Read(file2) → Read(file3)
# Good: Parallel (single message, 3 tool calls)
Read(file1) | Read(file2) | Read(file3)
# Bad: Heavy work in main thread (tokens visible)
[read 10 files, analyze, generate report]
# Good: Subagent isolation (only summary returned)
Task("Analyze 10 files, return 3-line summary")
# Bad (47 tokens)
"I'll help you implement that feature. Let me start by
examining the codebase to understand the current architecture,
then I'll make the necessary changes..."
# Good (3 tokens)
[starts making changes]
| Operation | ~Tokens |
|---|---|
| Read 100 lines code | 400-800 |
| Grep results (10 matches) | 200-400 |
| Edit file | 100-300 |
| Skill load | 500-2000 |
| Task subagent prompt | 200-500 |
| Task subagent result | 100-500 |
| Web search result | 500-1500 |
| Mermaid diagram | 100-300 |
Before responding:
As MINUS (-1) validator:
Σ(generator_tokens) + Σ(validator_savings) ≡ 0 (mod 3)
# Analyze thread token usage
just cheapskate-analyze <thread-id>
# Estimate remaining budget
just cheapskate-budget
# Compress context
just cheapskate-compress
parallel-fanout - Efficient parallel dispatchtriad-interleave - Balanced token streamsfrustration-eradication - Don't waste tokens on frustrationThis skill connects to the K-Dense-AI/claude-scientific-skills ecosystem:
general: 734 citations in bib.duckdbThis skill maps to Cat# = Comod(P) as a bicomodule in the equipment structure:
Trit: 0 (ERGODIC)
Home: Prof
Poly Op: ⊗
Kan Role: Adj
Color: #26D826
The skill participates in triads satisfying:
(-1) + (0) + (+1) ≡ 0 (mod 3)
This ensures compositional coherence in the Cat# equipment structure.