v0.28.37 Release Notes

Released 2026-07-26

v0.28.36 -> v0.28.37

Summary

Subagent execution gains explicit reasoning control while runtime output handling becomes more predictable. Telegram and TUI paths now render, sanitize, and resume sessions consistently.

翻譯

子代理執行新增明確的 reasoning 控制，並改善執行期輸出處理的一致性。Telegram 與 TUI 流程現在能更穩定地渲染、清理輸出並恢復 session。

⚠️ Breaking Changes

ExecWithSubagent signature changed

The function now accepts an explicit reasoning argument between model and systemPrompt.
Before:

func ExecWithSubagent(ctx context.Context, task, sessionIDInput, model, systemPrompt string, excludedTools []string, parentSessionID string) (string, error)

After:

func ExecWithSubagent(ctx context.Context, task, sessionIDInput, model, reasoning, systemPrompt string, excludedTools []string, parentSessionID string) (string, error)

Migration:

// Old
exec.ExecWithSubagent(ctx, task, sessionID, model, systemPrompt, excludedTools, parentSessionID)
// New: preserve the previous model-selected reasoning behavior
exec.ExecWithSubagent(ctx, task, sessionID, model, "", systemPrompt, excludedTools, parentSessionID)
// New: explicitly select a reasoning level
exec.ExecWithSubagent(ctx, task, sessionID, model, "low", systemPrompt, excludedTools, parentSessionID)

翻譯

ExecWithSubagent 在 model 與 systemPrompt 之間新增 reasoning 參數。既有呼叫端可傳入空字串以保留原本由模型設定決定 reasoning 的行為，或改為明確指定 reasoning level。

Changes

BREAKING

• Add explicit reasoning control to subagent execution and dispatch configuration (@pardnchiu) [a86e9ed2]

翻譯

• 為子代理執行與 dispatch 設定新增明確的 reasoning 控制

FIX

• Fix Telegram formatting, TUI rendering, and subagent guidance (@pardnchiu) [2fa04f24]
• Fix response sanitization and Telegram session handling (@pardnchiu) [a66680c2]

翻譯

• 修正 Telegram 格式、TUI 渲染與子代理指引
• 修正回應清理與 Telegram session 處理

Scope

• internal/agents/exec/ — BREAKING, FIX
• internal/tools/subagent/ — BREAKING, FIX
• internal/runtime/telegram/ — FIX
• internal/runtime/tui/ — FIX
• configs/prompts/ — BREAKING, FIX
• cmd/app/ — BREAKING

---

Generated by SKILL