docs(changeset): simplify print mode stream-json note

This commit is contained in:
Kaiyi 2026-06-29 19:41:44 +08:00
parent ebcaa405d6
commit 98f4cfd83d

View file

@ -2,4 +2,4 @@
"@moonshot-ai/kimi-code": patch
---
Expand `kimi -p` print mode to cover (and slightly exceed) kimi-cli's non-interactive `stream-json` capabilities: emit model thinking as its own JSONL line; add `--input-format text|stream-json` (multi-turn prompts over stdin), `--final-message-only` and the `--quiet` shorthand; surface background-task/cron events as `notification` lines; report turn failures as `{"type":"error",...}` JSON with retryable provider errors mapped to exit code 75; wait for background tasks before exit (gated by `background.keepAliveOnExit`); and emit the TUI activity layer (tool progress, subagent lifecycle, warnings, skill/MCP/compaction/goal/agent-status/tool-list updates) on stdout so the stream-json output stays both complete and entirely JSON.
Expand `kimi -p` print mode: add `--input-format text|stream-json` (multi-turn prompts over stdin), `--final-message-only`, and the `--quiet` shorthand. In `stream-json` mode the output is now entirely JSON and covers the full activity stream — model thinking, tool progress, notifications, subagent and lifecycle events, and turn errors (retryable provider errors map to exit code 75). Background tasks are drained before exit, gated by `background.keepAliveOnExit`.