skills.rest
amanpal3
amanpal3COMMUNITY
Explore all skills by this author →
GitHub

token-budget

Estimates and tracks token usage to prevent LLM context window overflow.

1|Updated Sep 3, 2026
Search Tags:#context management#agent workflow#progressive loading#context window#llm optimization#token budget#token estimation
One-click install
npx skills add https://github.com/amanpal3/SKILLs --skill token-budget-amanpal3
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: token-budget
Source: https://github.com/amanpal3/SKILLs/tree/main/.agent/plugins/gsd/.agents/skills/token-budget
Command: npx skills add https://github.com/amanpal3/SKILLs --skill token-budget-amanpal3

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Long agent sessions silently degrade as the context window fills up, causing quality loss and truncated outputs. This Skill gives agents a concrete method to estimate token consumption, monitor budget thresholds, and adjust loading strategy before context quality collapses. ## Core Features & Use Cases - Token Estimation Tables: Provides per-line token estimates for code, Markdown, JSON/YAML, and comments, plus file-size categories that dictate whether to load fully, outline, or search first. - Budget Threshold Protocol: Defines four quality tiers (PEAK, GOOD, DEGRADING, POOR) mapped to context usage percentages, with specific actions and alerts at 50% and 70% usage. - Optimization Strategies: Implements progressive loading levels, just-in-time file loading, search-before-load, and summarize-and-reference patterns to minimize token spend. - Use Case: During a multi-file refactoring task, the agent tracks cumulative tokens in a Token Tracker table, switches to outline-only loading at 50% budget, and triggers a state dump with /pause at 70% to hand off to a fresh session. ## Quick Start Ask the agent to estimate the current token budget and recommend whether to load a large file fully or search it first.

Frequently Asked Questions about token-budget

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I estimate token usage for a code file before loading it?▼

Multiply the file's line count by roughly 4 tokens per line for code, or 5-7 for JSON/YAML. Files under 50 lines can load freely, while files over 500 lines should never be loaded fully and require search or outline-based access.

How to prevent context window overflow in long AI agent sessions?▼

Track cumulative tokens per task in a budget table and follow threshold tiers: proceed normally under 30%, switch to search-first at 30-50%, use outlines only at 50-70%, and trigger a state dump with session handoff above 70%.

What is progressive loading for LLM context management?▼

Progressive loading retrieves context in stages: function signatures first, then key functions relevant to the task, then related code, and the full file only when essential. This avoids spending tokens on irrelevant file content.

When should an AI agent avoid loading a full file?▼

Avoid full loads for files over 200 lines when a targeted snippet suffices, and never fully load files over 500 lines. Search for relevant terms first, then load only the sections the current task requires.

What are the limitations of line-based token estimation?▼

Line-based estimates are approximations that vary with code verbosity, comment density, and language syntax. They guide loading strategy but cannot replace actual tokenizer counts for precise budget enforcement.

Related Skills

Explore More Software Engineering Skills→
LuuKhoaHocLuuKhoaHoc

Context Fetch

Search codebases for keywords before loading files to reduce context usage.

Community
Intermediate
ngocsangyemngocsangyem
14

meow:session-continuation

Persist and synchronize workflow state across sessions using TOON-based state format.

Community
Advanced
lucasFR2lucasFR2

context-compression

Optimizes AI agent context compression using Python scripts and Markdown documentation.

Community
Advanced
dotlab-hqdotlab-hq

langchain-rag

Build RAG pipelines that ingest, chunk, embed, and retrieve documents for LLMs.

Official
Intermediate