Context Doctor

扫描 OpenClaw 工作区引导文件与已安装技能，按字符/4 估算 token，可视化系统提示、工作区文件、技能列表和工具 Schema 的预算占比，给出截断、缺失和引导开销过高等诊断信号，可输出终端文本、JSON 或 PNG 图片。

jzocb

热度 96.42

最近更新 2026-05-11 17:29:37

分类：数据依赖：代码执行 / 读写文件应用场景：报告撰写 / 数据处理

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add context-doctor

Metadata

name: context-doctor
description: Visualize and diagnose OpenClaw context window usage. Generates a terminal-rendered breakdown showing workspace files (status, chars, tokens), installed skills inventory, and token budget allocation across bootstrap components. Use when: (1) user asks about context window health or token usage, (2) debugging agent quality degradation ("agent got dumber"), (3) after editing workspace files to verify impact, (4) auditing bootstrap overhead. NOT for: conversation history analysis, model selection, or cost tracking.

Context Doctor

Diagnose and visualize how the context window budget is allocated.

Quick Start

Run the visualization script:

python3 scripts/context-doctor.py

The script auto-detects:

Workspace path (~/.openclaw/workspace or OPENCLAW_WORKSPACE env)
OpenClaw version (via openclaw --version)
Installed skills (system + workspace)
Bootstrap file sizes and status

Output Sections

1. Workspace Files

Shows each bootstrap file with:

Status: ✓ OK / ✗ MISSING / ⚠ TRUNCATED (symlink target missing = MISSING)
Characters & Tokens: Raw size and estimated token count (chars/4)
Visual bar: Proportional to file size

2. Installed Skills

Lists all discovered SKILL.md files from system and workspace skill directories. Skills are loaded on-demand, not in bootstrap — shown for awareness only.

3. Token Budget

Breaks down bootstrap overhead:

System Prompt (framework)
Workspace Files (from scan)
Skills List (metadata descriptions)
Tool Schemas + Summaries

Shows total bootstrap % vs free conversation space.

Diagnostic Signals

Signal	Meaning	Action
🔴 TRUNCATED	File exceeded max chars, silently cut	Reduce file size or split content
🟡 MISSING	File not found or broken symlink	Check path, fix symlink target
🟢 Bootstrap <10%	Healthy	No action needed
🟠 Bootstrap >15%	Heavy	Review workspace files for bloat

Image Output (for chat / sharing)

Generate a PNG image directly — no terminal screenshot needed:

python3 scripts/context-doctor.py --png /tmp/context-doctor.png

The script renders a colored terminal-style image via Rich SVG export. Requires: rich (pip3 install rich) + one of: rsvg-convert (brew install librsvg) or cairosvg (pip3 install cairosvg).

Agent workflow: When a user asks about context health in chat, run with --png and send the image directly. The output path is printed to stdout on success.