Image Analyzer Vision

For text-only Token Plan models, use this skill to gain visual understanding by delegating to a vision-capable model (qwen3.7-plus, qwen3.6-plus, qwen3.5-plus, kimi-k2.5, qwen3.8-max).

User request: $ARGUMENTS

When to use

Recommended: switch to a vision model directly

The simplest path — if the user is on a text-only model but frequently needs vision, switch:

# Claude Code
/model qwen3.7-plus   # or qwen3.6-plus / qwen3.5-plus / kimi-k2.5

# OpenCode
/models   # then search and select

# Qwen Code
/model    # then select

# OpenClaw — edit ~/.openclaw/openclaw.json:
#   agents.defaults.model.primary → "bailian/qwen3.7-plus"

Then pass images directly: paste, drag-drop, or reference a path in the conversation.

Alternative: add this Skill for delegation

Use this only when the user must stay on the text-only model (e.g., code-generation specific model).

For Claude Code:

mkdir -p .claude/skills/image-analyzer

Create .claude/skills/image-analyzer/SKILL.md:

---
name: image-analyzer
description: Helps models without vision capabilities understand images. Use this skill when you need to analyze image content, extract information, text, or UI elements from images, or understand screenshots, charts, architecture diagrams, or any visual content. Simply pass in the image path to get a description.
model: qwen3.7-plus
---

qwen3.7-plus has visual understanding capabilities. Use qwen3.7-plus directly for image understanding.

When given an image path, read the image (you can pass it via the Read tool), then call qwen3.7-plus with the image to extract a description. Return the description to the user.

For OpenCode, instead use an Agent at .opencode/agents/image-analyzer.md:

---
description: Helps models without vision capabilities understand images.
mode: subagent
tools:
  bash: true
  write: false
  edit: false
---

Delegate to qwen3.7-plus (vision-capable) for image understanding. Pass the image path or URL via the prompt and request a structured description.

For OpenClaw / Hermes / Codex / Qwen Code / Qoder — see token-plan-bailian-config skill for the skill-paths table.

Steps

1. Identify the user's current model (from the active tool's model picker or ~/.openclaw/openclaw.json).
2. If vision-capable → tell them they already have it; do nothing.
3. If text-only → recommend switching to qwen3.7-plus (simplest), OR install the Skill/Agent pattern above (preserves current model).

Notes

• Cost: Switching to qwen3.7-plus is the same Token Plan Credits as the text-only model — no surcharge for vision. Using a Skill that delegates to qwen3.7-plus counts as two Token Plan invocations (parent + delegate).
• F2 TRUTH: never claim to "see" an image if your active model is text-only without this skill loaded. Be honest: "I'm on , switch to or load image-analyzer-vision first."
• OpenClaw cache invalidation: after editing ~/.openclaw/openclaw.json, clear ~/.openclaw/agents/main/agent/models.json and restart the gateway, otherwise old config sticks.