flow-skill-analyze-context

name	flow-skill-analyze-context
description	Analyze total token usage including conversation history, system prompts, and active rules to estimate cost.

Analyze Context (Total Cost Estimation)

Instructions

1. Scope: Analyze ALL components that contribute to the context window and cost. This includes:

   • Conversation History: User messages, Assistant responses, Tool calls, and Tool outputs.
   • System Context:
     • System Prompts: The core instructions defining the agent's behavior.
     • Active Rules: Content from AGENTS.md and any active .cursor/rules.
     • Skill/Tool Definitions: The descriptions of available tools and skills (often a significant overhead).
     • Attached Context: Any <open_and_recently_viewed_files>, <git_status>, or other automatic context blocks.

2. Estimation Method:

   • Calculate the token count based on the number of characters for each component.
   • Formula: Token Count = Character Count * Multiplier
   • Multiplier: Use 0.3 (approx. 1 token per 3.3 characters).

3. Procedure:

   1. Conversation: Estimate chars in the visible chat history.
   2. System/Rules: Read AGENTS.md and any active rule files to estimate their size.
   3. Overhead: Add a heuristic buffer for Tool/Skill definitions and hidden system prompts (typically 2000-5000 chars or ~1000 tokens depending on the number of tools).
   4. Total: Sum all components.
   5. Report: Provide a breakdown of the estimated usage.

4. Tools:

   • Use the script for calculation:

     deno run scripts/count_tokens.ts "TEXT_snippet"
     

   • Or manual calculation: Total Chars * 0.3.

Examples

User: "How many tokens are we using?" Action:

1. Chat: ~5000 chars (User + AI messages).
2. System: AGENTS.md (~3000 chars) + Rules (~1000 chars).
3. Overhead: ~3000 chars (Tool definitions).
4. Total Chars: 12,000.
5. Calc: 12,000 * 0.3 = 3,600 tokens.
6. Report: "Estimated Context Usage: ~3,600 tokens
   • Conversation: ~1,500 tokens
   • System & Rules: ~1,200 tokens
   • Tool Definitions: ~900 tokens"