dsh-subagent-vision

Overview

dsh bundle: subagent_vision — delegate image reading to a vision-capable model from a text-only session, plus paste-to-path so pasted images reach the subagent as file paths.

README / EN

Package documentation

Registry summary

dsh bundle: subagent_vision — delegate image reading to a vision-capable model from a text-only session, plus paste-to-path so pasted images reach the subagent as file paths.

dsh.pub verifies the pinned bundle contract, runtime facts, and distribution semantics. The complete README remains in the source repository.

Read the full README on GitHub

LIMITATIONS

Known limitations

- **No per-call model selection**: the child route is the deployment-configured `agentOptions`, not a tool argument the model can change mid-task. (This mirrors the stock subagent tools' fixed-route constraint.) - **Send-time conversion is all-or-nothing per send**: a draft's images are all converted before the send; a failed upload aborts the send (the composer restores the draft and thumbnails, so nothing is lost). - **Temp files accumulate**: pasted images land in the OS temp directory under `subagent-vision-paste-*`; nothing deletes them (the OS temp cleaner will). - **One-shot children**: the child settles and its conversation is read-only afterwards (use the stock `subagent`/continuable tools if you need to keep talking to a child). - **The parent model never sees the image itself** — only the child's text reading of it.

dsh-subagent-vision