v0.3.70

18 commits since v0.3.69. 57 files changed, +1310 / -128 lines.

✨ Features

• 822d7ef8 Kimi reasoning-effort config: per-model reasoning-effort field, validated against the model's allowed set
• 6295f8e8 Kimi plan-tier context window: API plan vs Coding plan drives the context window automatically
• fb89940e Skills review_gate: frontmatter declaration for high-stakes skills that require user approval before side effects
• 17cb7dae Plan discard autonomy gate: agent-initiated plan discard now requires plan autonomy, user-initiated discard always allowed
• 3eb6ab7d Plan yolo scope: design-track refusal scoped to /plan slash origin, not every yolo session

🔧 Fixes

• 90add3ca MiniMax live model fetch: fetch models from /v1/models endpoint instead of hardcoded list
• 3d805011 Kimi context_window persist: derived context_window saved to config on plan pick
• 584f14ca Moonshot config_for(): session slot resolves the moonshot provider correctly
• 6d640783 Moonshot keys.toml merge: provider key merged from keys.toml into the active config
• 9d85ca5f Compaction context window: honor the provider's configured window without a manual swap
• cd77d138 Phantom detector language-agnostic: accents and non-ASCII no longer disable strict detection
• 457f646a RSI proposals prune: already-handled entries removed from pending proposal files
• 4ef3af52 Mission-control Flakiest panel: windowed so fixed tools age out of the view
• d1abefc7 Phantom leading-imminence: catch work announcements like '... Now downloading X'

📖 Documentation

• 604f4226 Document skill review_gate and the yolo design-gate caveat
• 6c34134f Amend yolo Editing rule, document review gates

🧹 Miscellaneous

• 268140c9 perf(lazy-tools): tool_search activates on use, not all 8 matches (#604)
• 50004329 perf(lazy-tools): LRU-bound the per-session active tool set (#603)

📊 Stats

• 18 commits since v0.3.69
• 57 files changed, +1310 / -128 lines
• 5017 tests (5012 passed, 0 failed, 29 ignored)

Full Changelog: adolfousier/opencrabs@v0.3.69...v0.3.70