v0.6.4

zall 0.6.4

v0.6.3

feat(memory): 跨会话 /memory + 上游窗口事实化 + 选择器过滤 + 等待 spinner 修复 — 0.6.3

- /memory 命令 (user_profile/project_knowledge/error_patterns/decisions),
  记忆注入新会话 system prompt; 修复 Windows 下记忆从未落盘 (临时文件名
  含非法字符, mkstemp 修复)
- 上游 /models 即事实: 列表只列 live 模型 (未配 provider 折叠成一行),
  context_length 注入窗口表优先于内置猜测 (sensenova 网关 1M 不再显示 32k),
  REPL 启动后台探测; 窗口未知时只显已用 token 不伪造百分比
- /model 选择器: 打字即过滤, 无匹配可自由输入模型名 (字符校验防命令污染),
  降级单行路径共用契约 (ptk 菜单返回 (is_text, value))
- spinner 定位空行: 盲文帧缺字形 → 句点帧 (ascii/unicode 自适应)
- 测试: live window 不变量, 选择器过滤不变量, conftest 隔离真实全局表
  (_CUSTOM_WINDOWS/_LIVE_WINDOWS 每测重置, 探测线程 no-op)

v0.6.2

chore: release 0.6.2

v0.4.10 — interaction bug fixes + self-evolution wiring

v0.4.9 — Interaction reliability fixes

• A1: Stream exception silent swallow — now logs+propagates instead of silent truncation
• A2: CLI retry step count drift — new AgentLoop.retry_step() without counter increment
• A3: Spinner persistent thread destroyed — persistent thread, shutdown_spinner() for cleanup
• A4: Tool fold output unbounded — capped at 64, LRU eviction
• D: ContextLimitExceeded deprecated (backward-compat preserved)

v0.4.10 — Self-evolution / memory wiring

• B1: /suggest and /learn commands + auto-apply high-confidence adjust_k + config layer bridge
• B2: Cross-session learned memory (load_learned_memo) injected into system prompt

Tests

• 469 passed, 15 new regression test cases