dsh-read-image v0.1.0

dsh-read-image v0.1.0

Plug-and-play image reading for text-only DeepSeek Harness models — no preset changes required.

Features

• Pasted images are no longer rejected on text-only routes (admission gate declared open for image input)
• [Image #N] projection — image blocks are replaced in place with [Image #N] text at the model boundary; pixels never reach a text API
• First-class read_image tool, auto-registered per session (shadows the built-in tool):
  • image_index / file_path / prompt / reasoning_effort / timeout_ms / max_tokens / max_thinking_tokens
  • text-only routes: the configured vision model returns a text description; native multimodal routes get the image itself
• Native multimodal routes fall back to the harness's built-in read_image (no misleading [Image #N] section)
• Visual configuration — Settings → Read image page (typert Remote bridge, works around the harness's plugin-namespace settings allowlist)
• Environment self-check (fail-safe) — if any harness contract is missing on upgrade, the plugin loads nothing and the harness boots normally; diagnostics go to ~/.dsh/logs/dsh-read-image-guard.log

Install

dsh plugin --profile web add github:OoWJZZoO/dsh-read-image

then restart dsh web. Manual install instructions are in the README.

Compatibility

Built and validated against DeepSeek Harness 0.1.0-rc.6. The harness is in developer preview — a newer release candidate may need a compatibility pass (the plugin fails safe in that case).