Автономность мелких моделей: OUTPUT DISCIPLINE, milestone-todo, перехват финала хода ...

Два последовательных захода над одной проблемой — мелкие модели (12–30B) в navi_code
теряют автономность на трудных шагах: «остановился поболтать» вместо действия и
слишком большое расстояние между пунктами плана.

Заход 1 (З1–З3):
- З1: OUTPUT DISCIPLINE в системном промпте navi_code — «act, don't announce»,
  без few-shot антипримеров. Контракт хода: ответ без tool_calls = конец хода, поэтому
  объявление намерения текстом убивает автономность; правила заставляют вызывать
  инструмент в том же ходе.
- З2: плоский todo + метка группы milestone + декомпозиция. _parse_plan_steps
  возвращает list[tuple[milestone, text]]; milestone — метка группировки (не сущность,
  без статуса), «done» вычисляется при рендеринге; подшаги = больше плоских шагов
  (без вложенности). TUI side-panel группирует по milestone (плоский фолбэк при пустом
  milestone). Plan depth: max 15→20 + правило декомпозиции.
- З3: adaptive re-plan «длинный шаг» — nudge «разбей шаг» при in_progress ≥ порога
  итераций без смены todo (порог 4, раньше общего anti-stall warning на 8).

Заход 2 (шаги 1–3, после cloud-теста 31b vs 12b):
Корневая структурная причина: весь спасательный механизм (anti-stall warning с явным
предложением reflect, adaptive re-plan) живёт только внутри tool-цикла — nudge
инжектируется в pre_turn *следующей* итерации, которой при «остановился поболтать»
нет (ход закрылся по return до post_turn). 31b застревала через «продолжаю
tool-итерации» → дожала до warning → спаслась; 12b — через «замолчала текстом» → мимо
всех nudge.

- Шаг 1: перехват финала хода. Если модель выдала bare-text, но в todo есть открытые
  шаги (pending/in_progress) и лимит не исчерпан — НЕ эмитить StreamEnd, а сохранить
  ассистентский текст в session.context, поставить системный nudge и continue
  (без StreamEnd, без workers — консистентно с multi-iteration tool-турами). Счётчик
  final_interceptions на AgentTurnContext, лимит final_intercept_limit (default 2),
  эскалация жёсткости (мягкий → «second stop»). has_open_steps в todo.py: пустой
  todo → False (защита casual-сообщений), failed/skipped терминальны. Профильные
  флаги final_intercept_enabled/limit в base.py + loader + admin.
- Шаг 2: жёсткий reflect-триггер в промпте — «~3 tool attempts on the same step
  without progress → call reflect IN THIS TURN (tool call, not reasoning aloud)».
- Шаг 3: открыть replan для застревания — «call replan when reflect showed the whole
  approach is dead (not one failed step, but the approach itself)».

Тесты: 874 passed, 1 skipped. Новые — has_open_steps (5), final intercept (5),
milestone-группировка, adaptive long-step nudge, парсер шагов с milestone-маркером.

cloud: navi_code model → gemma4:31b-cloud для тестирования догадки (31b признала
застревание, 12b — нет); .env cloud-host уже gitignored.

Co-Authored-By: Claude <noreply@anthropic.com>

replan: integrated mid-task re-planning tool ...

Add a replan tool that re-runs the planner over the live session context
+ todo + scratchpad when the plan's structure is stale due to discoveries
(NOT failed steps -- that stays [Adaptive re-plan]). Integrated approach:
PlanningEngine.run gains is_replan/replan_context (suppresses DIRECT
shortcut and observe-skip, frames Phase 1 as a revision); ReplanRunner
packs reason/goal/todo/findings/errors and captures PlanReady; the tool
is exposed to navi_code/developer/tool_developer via a
current_replan_runner ContextVar set per-iteration in run_stream (correct
after switch_profile). New plan replaces the todo. Lazy events import
breaks the navi.tools -> navi.core -> navi.tools cycle.

Isolate sub-agent todo + render live todo in TUI side panel ...

Phase 1 — sub-agent todo isolation (backend):
- Add current_todo_session_id ContextVar; subagent_runner scopes it to the
  sub-agent's ephemeral run id so its auto-populated plan and todo updates
  land in an isolated KV row instead of clobbering the parent session's todo
  (which the parent's goal-anchoring reads every iteration).
- todo._sid() and planning.set_tasks prefer current_todo_session_id; the
  parent run leaves it unset, so all existing todo consumers (anti-stall,
  goal anchor, get_progress_message) behave exactly as before.

Phase 2 — live todo in the TUI right column:
- TodoUpdated event + emit it from the agent loop after planning auto-populate
  and after each tool-execution turn.
- GET /sessions/{id}/todos reads the parent session's todo KV row (explicit
  user_id/session_id, optional injected kv).
- api.get_todos + TodoList/TodoPanel widgets: status-coloured markers
  (pending dim, in_progress accent+bold, done success, failed error, skipped
  dim), progress header, scrollable panel below the auto-height info block.
- Hybrid delivery: REST seeds the panel on attach/switch, todo_updated WS
  events update it live.

Sub-agent todos as nested sub-lists is deferred to a later phase.

Co-Authored-By: Claude <noreply@anthropic.com>

agent: bounded autonomy — scope boundary + observe-vs-act ...

navi_code had unwanted "free flight": an observe request ("look at a
directory") triggered the full Phase 3 plan with milestones + auto-todo,
and goal_anchoring then drove the agent to finish those steps, climbing
into sibling projects and executing milestone docs it found.

Two toggleable, default-off profile flags (on for navi_code):

- scope_boundary_enabled: injects a standing system message keeping the
  agent within the literally requested scope; forbids acting on
  discovered TODO/roadmap/milestone docs (report only).
- observe_skips_plan_enabled: Phase 1 classifies MODE: observe|act; an
  observe request skips Phase 2/3 — no multi-step plan, no auto-todo, no
  "execute step by step" prompt. The agent just gathers info and answers.
  Independent of force_plan (observe on the first message still skips).

Free flight stays reproducible by flipping both flags off.

Co-Authored-By: Claude <noreply@anthropic.com>

agent: skip planning for casual greetings + strengthen DIRECT shortcut in Phase 1 ...

- Add fast _is_casual_message heuristic in navi/core/agent.py. Greetings and
  social chat (e.g. 'привет', 'как дела', 'hi', 'thanks') bypass planning even
  on the first session message, unless planning_mandatory is enabled.
- Strengthen Phase 1 planning prompt in navi/core/planning.py: explicitly
  require DIRECT output for greetings, simple questions, and one-step
  instructions.
- Fix broken ContextCompressed import in navi/core/compressor.py.
- Add unit tests covering the new heuristic and DIRECT prompt.

Co-Authored-By: Claude <noreply@anthropic.com>

Phase 2: Dual-write with is_context/is_display flags on Message ...

- Message model gets is_context and is_display bools
- PgSessionStore.save() writes flags directly to session_messages
- Agent sets is_context=False on display-only messages, is_display=False on context-only
- Planning: plan context msg is_display=False, plan marker is_context=False
- Compression: summarized messages get is_context=False, summary added to messages with is_display=False
- Tests updated for extra user display+context messages per turn

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Fix planning: change plan-follow-up role from system to user ...

After injecting the plan as an assistant message into session.context,
the previous code appended a system message saying "Plan is ready.
Execute it now..."  Many instruct-tuned models treat their own
assistant message as a completed response, and a trailing system
instruction is easy to ignore.

Changing the follow-up to role="user" makes the model see:
  assistant: plan
  user: "Execute this plan..."
which obligates the model to produce a new assistant response —
the tool-calling execution phase.

The follow-up message is appended only to session.context (LLM
context) and never to session.messages, so it is invisible in the
chat UI.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Enhance native toolset and add persistent KV store ...

- Add PostgreSQL-backed KvStore (navi/store/) for session-scoped data.
- Migrate todo and scratchpad from in-memory dicts to KvStore.
- Filesystem: add copy, grep, diff actions; compress description.
- CodeExec: remove language param, expose working_dir in schema.
- ImageView: resize to 1024px JPEG + Content-Type guard for URLs.
- Memory list: return distinct categories instead of all facts.
- SSH: add scp action with upload/download support.
- Update CLAUDE.md (Postgres-only), docs/tools.md, add docs/store.md.
- Fix agent/planning/context_builder async signatures for todo helpers.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Refactor: move tool helper modules into _internal subpackage ...

Moves non-tool infrastructure out of navi/tools/ root so that only
actual tool classes live there:

  base.py           → _internal/base.py
  loader.py         → _internal/loader.py
  middleware.py     → _internal/middleware.py
  logging_middleware.py → _internal/logging_middleware.py
  _time_parser.py   → _internal/time_parser.py

All imports updated across core/, api/, mcp/, tools/, and tests/.
No proxy files remain in navi/tools/ root.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Clarify knowledge persistence prompts

Integrate proactive knowledge capture into planning and persona ...

- Phase 1 planning: add KNOWLEDGE CAPTURE section to prompt
- Phase 3 planning: add knowledge persistence rule with target selection
- Persona: add "search before save" deduplication instruction
- MCP gnexus-book: add persistence guidance for discovered facts

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Add deterministic line-based file editing (edit_lines), rating UI fix, and session refresh ...

- filesystem.py: add edit_lines action (deterministic line ops via operations array)
           + numbered param for read (1-based line numbers in output)
           + clarify four editing modes in tool description
- chat.js: fix rating IDs for streaming messages (assign h_ ID on stream_end)
- SessionList.vue: mobile pull-to-refresh with PTR_THRESHOLD=80
- AppSidebar.vue: desktop refresh button next to Conversations header
- planning.py: knowledge source assessment in Phase 1
- debug panel: MCP servers tab + resolved tools per profile
- NAVI.md: reposition as neutral quick-reference

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Improve 3D modeling validation prompts

Open artifact links externally on Android

Extract PlanningEngine, ContextBuilder, ToolExecutor from agent.py ...

- navi/core/planning.py: new 3-phase planning pipeline (~390 lines)
- navi/core/context_builder.py: system prompt caching, memory/context injection, goal anchoring (~160 lines)
- navi/core/tool_executor.py: tool execution with middleware chain (~150 lines)
- navi/core/agent.py: reduced from ~1420 to ~770 lines; delegates to extracted classes

All compilation verified.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>