OpenKyrozen v2.0.4

What's Changed

• fix: keep self-learning alive after CLI exit by @EvanProgramming in #159
• fix: harden in-place update command by @EvanProgramming in #161
• fix: persist repeated self-learning preferences by @EvanProgramming in #163
• fix: start self-learning worker from TUI by @EvanProgramming in #165
• fix: eliminate Bubble Tea stream repaint latency by @EvanProgramming in #167
• fix: isolate interaction state by project scope by @EvanProgramming in #174
• fix: materialize accepted plan tasks exactly once by @EvanProgramming in #175
• fix: bind Agent actions to the accepted plan by @EvanProgramming in #176
• fix: normalize DeepSeek controls and sanitize streams by @EvanProgramming in #177
• fix: display effective Agent mode during plan execution by @EvanProgramming in #178
• test: gate the complete plan-to-deployment workflow by @EvanProgramming in #179
• fix: stop plan-mode execution loops by @EvanProgramming in #181
• fix: keep accepted plans authoritative by @EvanProgramming in #183
• fix: recover Markdown plan proposals by @EvanProgramming in #185
• fix: sanitize provider plan metadata by @EvanProgramming in #187
• fix: bind server lifecycle commands to plans by @EvanProgramming in #189
• fix: parse DeepSeek DSML tool calls by @EvanProgramming in #191
• fix: parse DeepSeek XML action blocks by @EvanProgramming in #193
• fix: renew capabilities during active turns by @EvanProgramming in #195
• docs: refresh discovered test count by @EvanProgramming in #197
• fix: recover plan after rejected provider action by @EvanProgramming in #199

Full Changelog: v2.0.3...v2.0.4