Prompt Compressor for Document Context

Prompt compressor for document context

Share the material you want to check and tell your AI what you want to know. It can pick out useful passages for you to compare with the original.

Use with your AI

For tasks with several steps, continue in the AI you already use.

CodexClaude CodeCursorWindsurfCline

View instructions

Use context-window-optimization-strale to find parts of the user’s material that may help with their question. Ask for the material or question if either is missing. If a link or file cannot be read, ask the user to paste its text. For multiple documents, keep them separate by giving each a short name yourself; do not ask the user to label or reformat them. Keep each selected part paired with its source and original wording. This tool only finds potentially useful text; it does not answer the question and may miss details. Compare the selection with the full source when available. Reuse the user’s Toolmu connection, or follow https://toolmu.com/SKILL.md to connect. Check current availability, price and the user’s budget before calling, and stay within the authorized amount.

Task:
From [the material I share], find parts that may help with [my question]. Tell me which document each part came from, and keep the original wording so I can check it. Do not answer the question for me.

Paste into your connected AI's conversation.

Connection guide

The final charge is based on confirmed usage. Unused reserved credits return to your available balance.

Strale

Prompt compressor for document context

Share the material you want to check and tell your AI what you want to know. It can pick out useful passages for you to compare with the original.

You receive

The tool picks out parts of your material that may help with the question. Compare them with the full material before relying on them; your AI can use the original to help answer.

Start with your task

From [the material I share], find parts that may help with [my question]. Tell me which document each part came from, and keep the original wording so I can check it. Do not answer the question for me.

What to know

It may miss useful details and does not answer the question by itself. Check the full material before relying on the result.

Inputs & outputs

Inspect the current inputs, availability, and maximum charge before starting.

Input schema

Input schema
Field / Path	Type	Requirement	Meaning
queryParams`queryParams`	`object`	Required	Required fields `documents, query`
Question or query`queryParams.query`	`string`	Required	The question or query used to judge which supplied chunks are relevant. State the information you want the context to help with.
Document chunks`queryParams.documents`	`array`	Required	The document chunks to consider for your question. The documented layout pairs an id with text; keep each identifier with the original content.
Requested token budget`queryParams.max_tokens`	`integer`	Optional	Optional token budget for the selection request. Provide your desired limit; the actual counting method and fit for a particular model need to be checked.

View raw JSON

{
  "type": "object",
  "required": [
    "queryParams"
  ],
  "properties": {
    "queryParams": {
      "type": "object",
      "required": [
        "documents",
        "query"
      ],
      "properties": {
        "query": {
          "type": "string"
        },
        "documents": {
          "type": "array",
          "description": "Array of objects with id and text fields."
        },
        "max_tokens": {
          "type": "integer"
        }
      }
    }
  },
  "additionalProperties": false
}

Result schema

Output fields have not been documented yet.

Additional notes

Strale’s capability description describes ranking and selecting document chunks by their relevance to a question, with a token budget in mind. Provide the chunks you want considered and the question you are working on. Review any selection, exclusions or scores the actual response supplies.

Tokens are units of text counted by a model when it processes input. A requested token budget is a planning limit; the counting method and intended model are not specified here. Before using any selected text, compare it with the source material for important omissions and check its token usage for your model. Keep any relevance scores under their supplied labels rather than assuming a scale or treating them as proof of answer accuracy.

Prompt compressor for document context

Prompt compressor for document context

You receive

Start with your task

What to know

Inputs & outputs

Input schema

Result schema

Connect the AI you use

Share the material and your question

Check the parts it found