Update Next Question to 0.2.0: click to fill composer #1267

Source

https://github.com/JacksonHuang517/paseo-next-question-plugin

Categories

• Utils
• Extras

Summary

Update the existing jacksonhuang517/next-question listing from 0.1.0 to 0.2.0. Requested Git revision: d89c38ba3b29da1d7d27f81b6c763b57f2de7849 (current main).

Clicking the suggested question or Use in composer now fills the original desktop/web conversation input. It appends after any existing draft, focuses the input, and does not send. Repeated clicks do not duplicate the same appended question. The existing Copy and inline Edit → Send question actions remain available, including on native mobile clients.

Paseo 0.13 has no public composer draft API. This update explicitly discloses a private desktop/web UI adapter in both overviews and README. It only accepts the standard MessageInput ref on the visible textarea's ancestor path, checks exact host + agent ownership and stale React alternate identity, rejects hidden/disabled/read-only/ambiguous inputs, and verifies the resulting text. Unsupported clients offer Copy/Edit. A server RPC revalidates the latest completed reply before insertion. Composer text stays on the current device and is not sent to prediction.

Verification

• 21 unit tests, TypeScript typecheck and Biome checks pass.
• Installed and running in the standard macOS Paseo 0.13.0 app; the user confirmed the click-to-fill interaction.
• Isolated real-React browser checks cover clickable text/button, append/preserve draft, empty draft, focus/send state, repeat clicks, no auto-send, stale reply, turn starting during validation, hidden-input fallback, inline explicit send, error/retry, and compact dark layout.
• Screenshots use the actual component in a sample-data harness, not a private transcript. Physical iOS/Android and a full real-app edge-case matrix remain untested.

Interface images

Requirements and data handling

Paseo >=0.13.0 <0.14.0 and an already configured Codex CLI on the daemon host. Existing prediction behavior remains: at most eight recent user/assistant messages (6,000 characters) sent to the configured Codex service, consuming the user's model allowance. Tool output, reasoning and attachment file contents are excluded. Ephemeral read-only Codex execution is not claimed to fully isolate the user's CLI configuration or every read-only tool. No credentials or transcript logging, runtime downloads or dependency-install/build scripts.

English/Chinese overviews and verification notes are updated.