Skip to content
TakoAPI
Back to skills

compaction-ui-enhancements

by maverick-software

Web & Frontend Development
352Downloads reported by clawskills.sh
1Views
0 likesSign in to like this skill
—Not yet rated
Your rating

Brief

Manual memory compaction button with context gauge for OpenClaw Control UI.

Description

What This Skill Does

Adds background memory compaction to the OpenClaw Control UI. Runs compaction without blocking the interface, triggers automatically when token usage exceeds a configurable threshold, and stores results with before/after token counts. Includes a settings tab with model selection and a context gauge button in the chat toolbar.

Runs entirely in the background via toasts instead of blocking the UI with a modal, so the chat remains interactive during compaction.

When to Use It

  • Compacting long agent sessions before context limit is hit
  • Reviewing how much memory was freed after a compaction run
  • Setting a custom model for compaction separate from the chat model
  • Automatically compacting after every response once tokens cross 80%
  • Reading the conversation summary paragraph in compacted output

Example Workflow

INPUT

User asks: compact this session automatically when it gets too full

AGENT
  1. 1

    User opens the Compaction settings tab under the Agent nav group

  2. 2

    Enables auto-compaction toggle and sets threshold slider to 75%

  3. 3

    After the next chat response, token usage crosses 75% and auto-compaction triggers

  4. 4

    A toast appears showing "Auto-Compacting (77%)" while the UI stays interactive

  5. 5

    Compaction completes and the Last Compaction card shows tokens before, after, and percent reduction

OUTPUT

Session memory compacted in the background; chat continues without interruption and result stored in compaction-config.json

Requirements

  • API key for any custom compaction model provider selected in the model selector

Installation

clawhub install maverick-software/compaction-ui-enhancements

Run the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.

Links