nanocode-cli 0.5.7__tar.gz → 0.5.10__tar.gz

{nanocode_cli-0.5.7 → nanocode_cli-0.5.10}/MANIFEST.in

@@ -1,3 +1,4 @@
 include LICENSE
+include README.zh-CN.md
 prune demo*
 prune tests

{nanocode_cli-0.5.7/nanocode_cli.egg-info → nanocode_cli-0.5.10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nanocode-cli
-Version: 0.5.7
+Version: 0.5.10
 Summary: A small terminal coding agent written in Python
 Author-email: hit9 <hit9@icloud.com>
 License-Expression: BSD-3-Clause
@@ -23,6 +23,7 @@ Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: anthropic>=0.64.0
 Requires-Dist: code-symbol-index>=0.1.13
+Requires-Dist: json-repair
 Requires-Dist: openai>=2.37.0
 Requires-Dist: prompt-toolkit>=3.0
 Requires-Dist: rich>=13.0
@@ -36,19 +37,22 @@ Dynamic: license-file
 
 A small terminal coding agent written in Python.
 
+[简体中文](README.zh-CN.md)
+
 nanocode is pre-1.0 software. Commands, configuration, and tool behavior may change before a stable release.
 
 ![nanocode screenshot](snapshots/nanocode-snapshot.png)
 
 ## Features
 
-- **Current turn flow**: Interim answers, tool results, and appended user input stay in order during a running task.
-- **Latest file state**: `Read` and `Edit` maintain a current, line-numbered file view with stale-range checks.
-- **Anchored edits**: `line:hash` anchors catch stale edits before they touch the wrong code.
-- **Working memory**: `Note` keeps the active goal, plan, and known facts separate from noisy tool output.
-- **Symbol index**: Jump from names to outlines, references, and changed files without searching blindly.
-- **Tool recall**: Prompt output is bounded, while raw `tr.N` results remain recallable when needed.
-- **Terminal-native UI**: Model picking, history search, confirmations, live command output, appended input, and status stay in the terminal.
+- **Live turn control**: Add follow-up input while the agent is still working, without losing the current tool flow.
+- **File-state brain**: Reads and edits build a current, line-numbered view of the files that matter now.
+- **Stale-edit protection**: `line:hash` anchors reject edits when the target code has drifted.
+- **Project-aware navigation**: Use the symbol index to jump through outlines, references, and changed files quickly.
+- **Recoverable context**: Tool output stays bounded in the prompt, while raw `tr.N` results remain recallable.
+- **Cache-aware context**: Stable sections stay early and noisy working state stays late to improve prompt-cache reuse.
+- **Focused working memory**: `Note` separates goal, plan, and known facts from noisy execution logs.
+- **Terminal-first workflow**: Model selection, history search, confirmations, live command output, appended input, and status all stay in one CLI.
 
 ## Install
 
@@ -56,6 +60,12 @@ nanocode is pre-1.0 software. Commands, configuration, and tool behavior may cha
 uv tool install nanocode-cli
 ```
 
+Upgrade:
+
+```sh
+uv tool upgrade nanocode-cli
+```
+
 For local development:
 
 ```sh

{nanocode_cli-0.5.7 → nanocode_cli-0.5.10}/README.md

@@ -2,19 +2,22 @@
 
 A small terminal coding agent written in Python.
 
+[简体中文](README.zh-CN.md)
+
 nanocode is pre-1.0 software. Commands, configuration, and tool behavior may change before a stable release.
 
 ![nanocode screenshot](snapshots/nanocode-snapshot.png)
 
 ## Features
 
-- **Current turn flow**: Interim answers, tool results, and appended user input stay in order during a running task.
-- **Latest file state**: `Read` and `Edit` maintain a current, line-numbered file view with stale-range checks.
-- **Anchored edits**: `line:hash` anchors catch stale edits before they touch the wrong code.
-- **Working memory**: `Note` keeps the active goal, plan, and known facts separate from noisy tool output.
-- **Symbol index**: Jump from names to outlines, references, and changed files without searching blindly.
-- **Tool recall**: Prompt output is bounded, while raw `tr.N` results remain recallable when needed.
-- **Terminal-native UI**: Model picking, history search, confirmations, live command output, appended input, and status stay in the terminal.
+- **Live turn control**: Add follow-up input while the agent is still working, without losing the current tool flow.
+- **File-state brain**: Reads and edits build a current, line-numbered view of the files that matter now.
+- **Stale-edit protection**: `line:hash` anchors reject edits when the target code has drifted.
+- **Project-aware navigation**: Use the symbol index to jump through outlines, references, and changed files quickly.
+- **Recoverable context**: Tool output stays bounded in the prompt, while raw `tr.N` results remain recallable.
+- **Cache-aware context**: Stable sections stay early and noisy working state stays late to improve prompt-cache reuse.
+- **Focused working memory**: `Note` separates goal, plan, and known facts from noisy execution logs.
+- **Terminal-first workflow**: Model selection, history search, confirmations, live command output, appended input, and status all stay in one CLI.
 
 ## Install
 
@@ -22,6 +25,12 @@ nanocode is pre-1.0 software. Commands, configuration, and tool behavior may cha
 uv tool install nanocode-cli
 ```
 
+Upgrade:
+
+```sh
+uv tool upgrade nanocode-cli
+```
+
 For local development:
 
 ```sh

nanocode_cli-0.5.10/README.zh-CN.md

@@ -0,0 +1,142 @@
+# nanocode
+
+一个用 Python 写的小型终端编程代理。
+
+[English](README.md)
+
+nanocode 仍是 1.0 前的软件。稳定版发布前，命令、配置和工具行为都可能变化。
+
+![nanocode screenshot](snapshots/nanocode-snapshot.png)
+
+## 特性
+
+- **实时回合控制**：代理还在工作时，也可以追加输入，不打断当前工具流程。
+- **文件状态大脑**：`Read` 和 `Edit` 会构建当前重要文件的带行号视图。
+- **过期编辑保护**：`line:hash` 锚点会在目标代码漂移后拒绝错误编辑。
+- **项目级导航**：通过符号索引快速查看 outline、references 和变更文件。
+- **可恢复上下文**：prompt 中的工具输出保持有界，原始 `tr.N` 结果仍可按需召回。
+- **缓存友好上下文**：稳定内容靠前，嘈杂的工作状态靠后，提高 prompt cache 复用率。
+- **聚焦工作记忆**：`Note` 把 goal、plan、known facts 从嘈杂执行日志中拆出来。
+- **终端优先工作流**：模型选择、历史搜索、确认、实时命令输出、追加输入和状态展示都在一个 CLI 内完成。
+
+## 安装
+
+```sh
+uv tool install nanocode-cli
+```
+
+升级：
+
+```sh
+uv tool upgrade nanocode-cli
+```
+
+本地开发：
+
+```sh
+uv sync --extra dev
+uv run nanocode
+```
+
+## 使用
+
+启动 CLI：
+
+```sh
+nanocode
+```
+
+常用参数：
+
+- `--config <path>`：使用指定 TOML 配置文件。
+- `--init-config`：创建默认配置文件。
+- `--yolo`：跳过会修改环境的工具确认。
+- `-v`, `--version`：显示版本。
+
+代理运行中，`+>` 提示符可以接收追加输入，并在下一次模型请求中发送。
+
+## 命令
+
+- `/help`：显示命令和工具。
+- `/status`：显示运行状态。
+- `/config`：显示当前配置。
+- `/api [auto|chat|anthropic]`：显示或设置 provider API 格式。
+- `/debug [on|off]`：切换模型 I/O debug trace。
+- `/compact`：立即压缩上下文。
+- `/index [force]`：同步或重建代码符号索引。
+- `/provider [NAME]`：显示或设置 provider。
+- `/model [MODEL]`：显示或设置模型。
+- `/reason`：选择 reasoning effort。
+- `/set KEY VALUE`：设置 provider/runtime 值。
+- `/yolo`：切换工具确认。
+- `/exit`, `/quit`：退出。
+
+交互选择器支持 `j`/`k`、方向键、`/` 搜索、Enter 和 Esc。输入框支持历史、补全和 `Ctrl-R` 历史搜索。
+
+## 工具
+
+- 文件：`Read`, `LineCount`, `List`, `Find`, `Search`。
+- 代码索引：`InspectCode`。
+- 编辑：`Edit` 创建或修改文件内容。
+- Shell：`Bash`, `Git`。
+- 工具结果：`Recall`。
+- 工作笔记：`Note`。
+
+`Read`、`Search` 和 `InspectCode` 会在合适时返回行锚点。`Edit` 使用当前 `line:hash` 锚点拒绝过期编辑。
+
+## 配置
+
+运行：
+
+```sh
+nanocode --init-config
+```
+
+默认配置位置是 `~/.nanocode/config.toml`。
+
+主要字段：
+
+- `[provider] active = "name"`
+- `[provider.<name>]`：`url`, `key`, `model`, `api`, `prompt_cache_key`, `available_models`, `reasoning`, `chat_reasoning`, `temperature`, `timeout`
+- `[paths] data_dir`
+- `[runtime] shell_timeout`, `max_agent_steps`, `max_context_tokens`, `yolo`
+
+`api = "auto"` 会根据 provider/model profile 在 Chat Completions 和 Anthropic Messages 之间选择。`prompt_cache_key = "auto"` 会根据 provider、model、workspace 和工具 schema 名称生成稳定 key。
+
+## 上下文设计
+
+每次模型请求都由 nanocode 手动构建成明确的 messages。稳定上下文在前，会话作为 messages 保留，工作记忆随后，最新文件状态放在末尾。
+
+```text
+model request
++--------------------------------------------------+
+| system                                           |
+|   concise agent contract and tool rules          |
++--------------------------------------------------+
+| user                                             |
+|   Environment                                   |
++--------------------------------------------------+
+| user/assistant                                  |
+|   conversation, compacted summaries, tools      |
++--------------------------------------------------+
+| user                                             |
+|   Memory: goal, plan, known, date               |
++--------------------------------------------------+
+| user                                             |
+|   FILE STATE: latest Read/Edit file view        |
++--------------------------------------------------+
+```
+
+核心规则：
+
+- 回合中的 assistant 文本和用户追加输入都会作为 conversation 保留。
+- 上下文过大时，较早 conversation 会压缩成明确的 summary。
+- FILE STATE 由成功的 `Read` 和 `Edit` 更新，展示当前文件范围，最近文件优先。
+- 更新的文件行会覆盖旧行；edit invalidation 会清理过期范围。
+- 文件行展示前会通过当前文件 stat 或行 hash 校验。
+- 成功的 `Read` 和 `Edit` 工具消息只指向 FILE STATE，不重复塞入文件正文。
+- 其他工具输出在 conversation messages 中保持有界，并可通过 `tr.N` 召回。
+
+## 安全
+
+nanocode 会在启动它的环境中编辑文件和执行 shell 命令。它不提供 sandbox 保护。需要隔离时，请在你自己的 sandbox、容器、虚拟机或其他隔离环境中运行。

{nanocode_cli-0.5.7 → nanocode_cli-0.5.10}/nanocode.py

@@ -11,10 +11,12 @@ import os
 import platform
 import re
 import selectors
+import shlex
 import shutil
 import signal
 import subprocess
 import sys
+import tempfile
 import threading
 import time
 import tomllib
@@ -28,6 +30,7 @@ from urllib.request import Request, urlopen
 
 import code_symbol_index as csi
 from anthropic import Anthropic
+from json_repair import repair_json
 from openai import OpenAI
 from prompt_toolkit import print_formatted_text, search as pt_search
 from prompt_toolkit.application import Application, run_in_terminal
@@ -52,7 +55,7 @@ from prompt_toolkit.widgets import SearchToolbar
 from rich.console import Console
 from rich.markdown import Markdown
 
-__version__ = "0.5.7"
+__version__ = "0.5.10"
 
 Json = dict[str, Any]
 HTTP_USER_AGENT = "nanocode/" + __version__
@@ -407,6 +410,7 @@ class AgentState:
     goal: str = ""
     plan: list[str] = field(default_factory=list)
     known: list[str] = field(default_factory=list)
+    check: str = ""
     summary: str = ""
     code_index_status: str = ""
     code_index_error: str = ""
@@ -439,7 +443,7 @@ class AgentState:
         return next((text for item in self.plan if (text := self.plan_text(item))), "")
 
     def apply(self, data: Json) -> None:
-        for attr in ("goal", "summary"):
+        for attr in ("goal", "summary", "check"):
             if isinstance(data.get(attr), str):
                 setattr(self, attr, str(data[attr]).strip())
         for attr in ("plan", "known"):
@@ -450,7 +454,7 @@ class AgentState:
 
     def format(self) -> str:
         known = ["- " + item for item in self.known] or ["- (empty)"]
-        return "\n".join(["Goal: " + (self.goal or "(empty)"), "Plan:", *self.plan_rows(), "Known:", *known])
+        return "\n".join(["Goal: " + (self.goal or "(empty)"), "Plan:", *self.plan_rows(), "Known:", *known, "Check: " + (self.check or "(empty)")])
 
 
 @dataclass
@@ -549,6 +553,7 @@ class Session:
     tool_counter: int = 0
     usage: ModelUsage = field(default_factory=ModelUsage)
     update: UpdateStatus = field(default_factory=UpdateStatus)
+    _gitignore_cache: dict[str, tuple[int, list[str]]] = field(default_factory=dict)
 
     def __post_init__(self) -> None:
         if self.system_info is None:
@@ -755,12 +760,23 @@ class Tool:
 
     def gitignore_patterns(self, root: str) -> list[str]:
         patterns = []
-        for path in dict.fromkeys([os.path.join(self.session.cwd, ".gitignore"), *([os.path.join(root, ".gitignore")] if os.path.isdir(root) else [])]):
+        cache = self.session._gitignore_cache
+        paths = [os.path.join(self.session.cwd, ".gitignore")]
+        if os.path.isdir(root):
+            paths.append(os.path.join(root, ".gitignore"))
+        for path in dict.fromkeys(paths):
             try:
+                mtime = os.stat(path).st_mtime_ns
+                cached = cache.get(path)
+                if cached is not None and cached[0] == mtime:
+                    patterns.extend(cached[1])
+                    continue
                 with open(path, encoding="utf-8") as file:
-                    patterns.extend(line.strip() for line in file if line.strip() and not line.lstrip().startswith("#") and not line.startswith("!"))
+                    pats = [line.strip() for line in file if line.strip() and not line.lstrip().startswith("#") and not line.startswith("!")]
+                cache[path] = (mtime, pats)
+                patterns.extend(pats)
             except OSError:
-                pass
+                cache.pop(path, None)
         return patterns
 
     def ignored(self, path: str, patterns: list[str]) -> bool:
@@ -1742,7 +1758,7 @@ class EditTool(Tool):
 
 class BashTool(Tool):
     NAME = "Bash"
-    DESCRIPTION = "Run one bash shell invocation in the workspace; returns exit_code/stdout/stderr and shows live output."
+    DESCRIPTION = "Run one bash shell invocation in the workspace; returns exit_code/stdout/stderr and shows live output. Avoid unbounded output; limit noisy commands with head/tail/sed/rg filters or command-specific limits, and inspect large outputs in chunks."
     SIGNATURE = "Bash(command)"
     EXAMPLE = (
         'Check environment. Example: {"command":"python3 --version"}',
@@ -1869,20 +1885,22 @@ class BashTool(Tool):
 
 class GitTool(Tool):
     NAME = "Git"
-    DESCRIPTION = "Run git argv in the workspace; returns exit/stdout/stderr, with approval for mutating commands."
-    SIGNATURE = "Git(argv=[...], cwd?)"
+    DESCRIPTION = 'Run git with explicit argv (default: cwd from Environment; use cwd= for other directories). For add, pass explicit file paths; broad add is rejected.'
+    SIGNATURE = "Git(argv=[command,...], cwd?)"
     EXAMPLE = (
-        'Read repo status. Example: {"argv":["status","--short"]}',
-        'Diff inside a subdir. Example: {"cwd":"src","argv":["diff","--","app.py"]}',
-        'Read history/object. Example: {"argv":["show","--stat","HEAD"]}',
+        'Status. Example: {"argv":["status","--short"]}',
+        'Diff. Example: {"cwd":"src","argv":["diff","--","app.py"]}',
+        'Stage explicit files. Example: {"argv":["add","--","nanocode.py","README.md"]}',
+        'Show commit/file. Example: {"argv":["show","--stat","HEAD"]}',
     )
     READONLY = {"status", "diff", "log", "show", "rev-parse", "ls-files", "grep", "blame"}
+    BROAD_ADD = {"-A", "--all", ".", "./", ":/", "*"}
 
     @classmethod
     def params_schema(cls) -> Json:
         return {
             "type": "object",
-            "properties": {"cwd": {"type": "string"}, "argv": {"type": "array", "items": {"type": "string"}, "minItems": 1}},
+            "properties": {"cwd": {"type": "string"}, "argv": {"type": "array", "items": {"type": "string", "minLength": 1}, "minItems": 1}},
             "required": ["argv"],
             "additionalProperties": False,
         }
@@ -1923,8 +1941,26 @@ class GitTool(Tool):
                 raise ToolError("git cwd is not a directory")
         if not args:
             raise ToolError("Git requires arguments")
+        self.validate_add(args, cwd)
         return args, cwd
 
+    def validate_add(self, args: list[str], cwd: str) -> None:
+        if args[0] != "add":
+            return
+        paths, explicit = [], False
+        for arg in args[1:]:
+            if arg == "--":
+                explicit = True
+            elif arg in self.BROAD_ADD or arg.startswith("--pathspec-from-file"):
+                raise ToolError("Git add requires explicit file paths")
+            elif explicit or not arg.startswith("-"):
+                paths.append(arg)
+        if not paths:
+            raise ToolError("Git add requires explicit file paths")
+        for path in paths:
+            if path.startswith(":") or any(char in path for char in "*?[]") or not self.session.in_cwd(os.path.abspath(os.path.join(cwd, path))):
+                raise ToolError("Git add requires explicit file paths inside workspace")
+
 
 class RecallTool(Tool):
     NAME = "Recall"
@@ -2013,15 +2049,15 @@ class RecallTool(Tool):
 
 class NoteTool(Tool):
     NAME = "Note"
-    DESCRIPTION = "Maintain durable working notes; goal and plan replace current values, known appends unique facts."
-    SIGNATURE = "Note(goal?, plan?, known?)"
-    EXAMPLE = ('Set memory. Example: {"goal":"ship parser fix","plan":["inspect parser","patch bug"],"known":["tests use pytest"]}',)
+    DESCRIPTION = "Maintain durable working notes; goal, plan, and check replace current values, known appends unique facts."
+    SIGNATURE = "Note(goal?, plan?, known?, check?)"
+    EXAMPLE = ('Set memory. Example: {"goal":"ship parser fix","plan":["inspect parser","patch bug"],"known":["tests use pytest"],"check":"pytest passed"}',)
     STORES_RESULT = False
 
     @classmethod
     def params_schema(cls) -> Json:
         strings = {"type": "array", "items": {"type": "string"}, "minItems": 1}
-        return {"type": "object", "properties": {"goal": {"type": "string"}, "plan": strings, "known": strings}, "additionalProperties": False}
+        return {"type": "object", "properties": {"goal": {"type": "string"}, "plan": strings, "known": strings, "check": {"type": "string"}}, "additionalProperties": False}
 
     @classmethod
     def payload_args(cls, payload: Json) -> list[Any]:
@@ -2031,7 +2067,7 @@ class NoteTool(Tool):
         if len(self.args) != 1 or not isinstance(self.args[0], dict):
             raise ToolError("Note requires named fields")
         data = self.args[0]
-        if unexpected := sorted(set(data) - {"goal", "plan", "known"}):
+        if unexpected := sorted(set(data) - {"goal", "plan", "known", "check"}):
             raise ToolError("Note unexpected field: " + ", ".join(unexpected))
         changed = []
         if "goal" in data:
@@ -2047,8 +2083,11 @@ class NoteTool(Tool):
                 raise ToolError("Note known must be an array")
             self.session.state.known = list(dict.fromkeys([*self.session.state.known, *(str(item).strip() for item in data["known"] if str(item).strip())]))
             changed.append("known")
+        if "check" in data:
+            self.session.state.check = str(data["check"]).strip()
+            changed.append("check")
         if not changed:
-            raise ToolError("Note requires goal, plan, or known")
+            raise ToolError("Note requires goal, plan, known, or check")
         return "Updated memory: " + ", ".join(changed)
 
     def short_args(self) -> list[str]:
@@ -2062,6 +2101,8 @@ class NoteTool(Tool):
             known = [Tool.compact(item, 120) for item in data["known"] if str(item).strip() and str(item).strip() not in self.session.state.known]
             if known:
                 lines.extend(["known:", *(f"  + {item}" for item in known)])
+        if check := str(data.get("check") or "").strip():
+            lines.append("check -> " + Tool.compact(check, 120))
         return ["\n".join(lines) or "{}"]
 
 
@@ -2091,6 +2132,11 @@ class ToolCall:
 class ContextManager:
     COMPACT_TITLE: ClassVar[str] = "--- Prior Conversation Summary (compacted) ---"
     COMPACT_RECENT_MESSAGES: ClassVar[int] = 8
+    CODE_EXTENSIONS: ClassVar[set[str]] = set(
+        ".c .cc .cpp .cxx .css .go .h .hpp .html .java .js .json .jsx .kt .lua .php .py .rb .rs .scss .sh .sql "
+        ".swift .toml .ts .tsx .vue .yaml .yml".split()
+    )
+    CODE_FILENAMES: ClassVar[set[str]] = {"CMakeLists.txt", "Dockerfile", "Makefile", "go.mod", "package.json", "pyproject.toml"}
 
     @dataclass
     class FileContextItem:
@@ -2123,28 +2169,35 @@ class ContextManager:
 
     def update_percent(self, messages: list[Json]) -> int:
         tokens = self.estimated_tokens(messages)
-        self.session.state.context_percent = min(100, round(tokens * 100 / self.session.settings.max_context_tokens))
+        self.session.state.context_percent = min(100, tokens * 100 // self.session.settings.max_context_tokens)
         return self.session.state.context_percent
 
     def maybe_compact(self, model: "ModelClient", base_system: str, turn_messages: list[Json] | None = None) -> None:
-        if self.estimated_tokens(self.model_messages(base_system, turn_messages)) < self.session.settings.max_context_tokens:
+        if not self.over_budget(base_system, turn_messages):
             return
         compacted, keep = self.compaction_parts()
-        if not compacted:
-            return
-        try:
-            self.apply_compaction(model.compact(self.compaction_input(compacted)), keep, turn_messages)
-        except Exception:
-            self.session.state.summary = (self.session.state.summary + "\nPrevious context was deterministically trimmed.").strip()
-            summary = self.session.state.summary
-            self.session.messages = ([{"role": "user", "content": self.COMPACT_TITLE + "\n" + summary}] if summary else []) + keep
-            self.prune_tool_records([*keep, *(turn_messages or [])])
+        if compacted:
+            try:
+                self.apply_compaction(model.compact(self.compaction_input(compacted)), keep, turn_messages)
+            except Exception:
+                self.apply_compaction_fallback(keep, turn_messages)
+        if turn_messages is not None and self.over_budget(base_system, turn_messages):
+            compacted, keep = self.turn_compaction_parts(turn_messages)
+            if compacted:
+                try:
+                    self.apply_turn_compaction(model.compact(self.compaction_input(compacted)), keep, turn_messages)
+                except Exception:
+                    self.apply_turn_compaction_fallback(keep, turn_messages)
+
+    def over_budget(self, base_system: str, turn_messages: list[Json] | None = None) -> bool:
+        return self.estimated_tokens(self.model_messages(base_system, turn_messages)) >= self.session.settings.max_context_tokens
 
     def memory_context(self, *, with_date: bool = False) -> str:
         rows = [
             "Goal: " + (self.session.state.goal or "(empty; use Note for multi-step work)"),
             "Plan:\n" + "\n".join(self.session.state.plan_rows() or ["- (empty; use Note for a short plan)"]),
             "Known:\n" + "\n".join("- " + item for item in self.session.state.known or ["(empty)"]),
+            "Check: " + (self.session.state.check or "(empty)"),
         ]
         if with_date:
             rows.append("Date: " + datetime.now().astimezone().strftime("%Y-%m-%d"))
@@ -2242,8 +2295,10 @@ class ContextManager:
             )
 
         paths = sorted((path for path in lines_by_path if lines_by_path[path]), key=lambda path: (-recent(path), path))
+        code_edits = self.recent_code_edits()
+        check_status = self.check_status(code_edits)
         focus, actions, errors = self.session.state.current_focus(), self.recent_file_actions(), self.recent_tool_errors()
-        if not paths and not omitted and not focus and not actions and not errors:
+        if not paths and not omitted and not focus and not actions and not code_edits and not check_status and not errors:
             return ""
         chunks = ["Read/Edit outputs update this section. Treat listed ranges as current file state."] if paths else []
         if focus:
@@ -2254,6 +2309,10 @@ class ContextManager:
                 chunks.extend(f"- {path} {start}:{end} current" for start, end in self.coverage(lines_by_path[path]))
         if actions:
             chunks.extend(["", "Recent file events:", *actions])
+        if code_edits:
+            chunks.extend(["", "Recent code edits:", *code_edits])
+        if check_status:
+            chunks.extend(["", "Check status:", *check_status])
         if errors:
             chunks.extend(["", "Recent tool errors:", *errors])
         if paths:
@@ -2267,7 +2326,6 @@ class ContextManager:
             chunks.append("Omitted content:")
             for path in sorted(omitted):
                 chunks.extend(f"- {path} source={source} lines={count}" for source, count in sorted(omitted[path].items()))
-        chunks.extend(["", "OUTPUT IN USER LANGUAGE"])
         return "\n".join(chunks).strip() if len(chunks) > 4 else ""
 
     def recent_file_actions(self) -> list[str]:
@@ -2284,6 +2342,30 @@ class ContextManager:
             for record in self.session.tool_errors[-5:]
         ]
 
+    def recent_code_edits(self) -> list[str]:
+        rows: dict[str, str] = {}
+        for record in self.session.tool_records[-20:]:
+            if record.name == "Edit":
+                for match in re.finditer(r'<Edit\s+path=(".*?")', record.output):
+                    try:
+                        path = str(json.loads(match.group(1)))
+                    except json.JSONDecodeError:
+                        continue
+                    if self.code_like_path(path):
+                        rows[path] = f"- {record.key} Edit {path}"
+        return list(rows.values())[-8:]
+
+    def check_status(self, code_edits: list[str]) -> list[str]:
+        if not code_edits:
+            return []
+        check = self.session.state.check.strip()
+        return ["- " + check] if check else ["- Code changed recently. Use Note(check=...) after checks, or final must say checks not run."]
+
+    @classmethod
+    def code_like_path(cls, path: str) -> bool:
+        name = os.path.basename(path)
+        return name in cls.CODE_FILENAMES or os.path.splitext(name)[1].lower() in cls.CODE_EXTENSIONS
+
     def coverage(self, numbered: dict[int, tuple[str, str, str]]) -> list[tuple[int, int]]:
         numbers = sorted(numbered)
         if not numbers:
@@ -2320,7 +2402,7 @@ class ContextManager:
         return segments
 
     def compaction_input(self, messages: list[Json]) -> str:
-        older, recent = self.compaction_recent(messages)
+        older, recent = self.compaction_parts_for(messages)
         return "\n\n".join(
             [
                 "State:\n" + self.session.state.format(),
@@ -2334,7 +2416,14 @@ class ContextManager:
         index = self.latest_user_index(self.session.messages)
         return (self.session.messages, []) if index is None else (self.session.messages[:index], self.session.messages[index:])
 
-    def compaction_recent(self, messages: list[Json]) -> tuple[list[Json], list[Json]]:
+    def turn_compaction_parts(self, messages: list[Json]) -> tuple[list[Json], list[Json]]:
+        index = self.latest_user_index(messages)
+        if index is None:
+            return self.compaction_parts_for(messages)
+        compacted, keep = self.compaction_parts_for(messages[index + 1 :])
+        return compacted, messages[: index + 1] + keep
+
+    def compaction_parts_for(self, messages: list[Json]) -> tuple[list[Json], list[Json]]:
         cut = max(0, len(messages) - self.COMPACT_RECENT_MESSAGES)
         return messages[:cut], messages[cut:]
 
@@ -2347,6 +2436,24 @@ class ContextManager:
         self.session.messages = ([{"role": "user", "content": self.COMPACT_TITLE + "\n" + summary}] if summary else []) + keep
         self.prune_tool_records([*self.session.messages, *(tool_messages or [])])
 
+    def apply_compaction_fallback(self, keep: list[Json], tool_messages: list[Json] | None = None) -> None:
+        self.session.state.summary = (self.session.state.summary + "\nPrevious context was deterministically trimmed.").strip()
+        summary = self.session.state.summary
+        self.session.messages = ([{"role": "user", "content": self.COMPACT_TITLE + "\n" + summary}] if summary else []) + keep
+        self.prune_tool_records([*keep, *(tool_messages or [])])
+
+    def apply_turn_compaction(self, data: Json, keep: list[Json], turn_messages: list[Json]) -> None:
+        self.session.state.apply(data)
+        summary = self.session.state.summary
+        index = self.latest_user_index(keep)
+        insert = len(keep) if index is None else index + 1
+        turn_messages[:] = keep[:insert] + ([{"role": "user", "content": self.COMPACT_TITLE + "\n" + summary}] if summary else []) + keep[insert:]
+        self.prune_tool_records([*self.session.messages, *turn_messages])
+
+    def apply_turn_compaction_fallback(self, keep: list[Json], turn_messages: list[Json]) -> None:
+        self.session.state.summary = (self.session.state.summary + "\nCurrent turn context was deterministically trimmed.").strip()
+        self.apply_turn_compaction({"summary": self.session.state.summary}, keep, turn_messages)
+
     def prune_tool_records(self, keep_messages: list[Json]) -> None:
         records = self.session.tool_records
         keep = set(re.findall(r"\btr\.\d+\b", self.messages_text(keep_messages)))
@@ -2629,6 +2736,7 @@ class ToolRunner:
         self.input_fn = input_fn
         self.output_fn = output_fn
         self.preview_fn: Callable[[str], bool] | None = None
+        self.preview_full_fn: Callable[[str], None] | None = None
         self.live_output: Callable[[str, str], None] | None = None
         self.live_start: Callable[[], None] | None = None
 
@@ -2783,8 +2891,11 @@ class ToolRunner:
         CodeIndex(self.session).update(list(dict.fromkeys(paths)))
 
     def confirm(self, call: ToolCall, tool: Tool, batch_suffix: str = "", planned_edit: EditBatchPlan.PlannedEdit | None = None) -> tuple[bool, str]:
-        if not (self.preview_fn and self.preview_fn(self.approval_display(call, tool, "confirm", batch_suffix=batch_suffix, planned_edit=planned_edit))):
-            self.output_fn(self.approval_display(call, tool, "confirm", batch_suffix=batch_suffix, planned_edit=planned_edit))
+        display = self.approval_display(call, tool, "confirm", batch_suffix=batch_suffix, planned_edit=planned_edit, preview_lines=40)
+        if self.preview_full_fn and tool.NAME == "Edit":
+            self.preview_full_fn(self.approval_display(call, tool, "confirm", batch_suffix=batch_suffix, planned_edit=planned_edit, preview_lines=None))
+        if not (self.preview_fn and self.preview_fn(display)):
+            self.output_fn(display)
         answer = self.input_fn("[Y/n or reason] ").strip()
         lower = answer.lower()
         if lower in {"", "y", "yes"}:
@@ -2792,19 +2903,26 @@ class ToolRunner:
         return False, "" if lower in {"n", "no"} else answer
 
     def approval_display(
-        self, call: ToolCall, tool: Tool, status: str, batch_suffix: str = "", planned_edit: EditBatchPlan.PlannedEdit | None = None
+        self,
+        call: ToolCall,
+        tool: Tool,
+        status: str,
+        batch_suffix: str = "",
+        planned_edit: EditBatchPlan.PlannedEdit | None = None,
+        preview_lines: int | None = 40,
     ) -> str:
         header = self.with_batch_suffix(("approve " if status == "confirm" else "auto ") + self.short_call(call), batch_suffix)
         if tool.NAME != "Edit":
             return header
         preview = planned_edit.preview(tool) if planned_edit and isinstance(tool, EditTool) else tool.preview()
-        return header + (("\n" + block) if (block := self.preview_block(preview)) else "")
+        return header + (("\n" + block) if (block := self.preview_block(preview, max_lines=preview_lines)) else "")
 
-    def preview_block(self, preview: str, *, max_lines: int = 40) -> str:
+    def preview_block(self, preview: str, *, max_lines: int | None = 40) -> str:
         lines = preview.rstrip().splitlines()
         if not lines:
             return ""
-        lines = lines[:max_lines] + (["... preview truncated ..."] if len(lines) > max_lines else [])
+        if max_lines is not None and len(lines) > max_lines:
+            lines = lines[:max_lines] + [f"... preview truncated: {len(lines) - max_lines} more lines (Ctrl-A: full preview) ..."]
         return "\n".join(["  preview", *("  " + line for line in lines)])
 
     def finish_display(
@@ -2981,7 +3099,8 @@ class ModelClient:
     def compact(self, context: str) -> Json:
         prompt = """
 Compact the nanocode working context.
-Return exactly one JSON object with keys: summary, goal, plan, known.
+Return one JSON object only. No markdown, prose, code fences, or comments.
+Use keys: summary, goal, plan, known.
 Rewrite recent conversation briefly inside summary.
 Keep only durable facts needed to continue; preserve file paths, symbols, constraints, and tr.N keys.
 """.strip()
@@ -2991,13 +3110,29 @@ Keep only durable facts needed to continue; preserve file paths, symbols, constr
             if self.session.config.provider.resolved_api() == "anthropic"
             else self.chat_request(messages, None, activity="compact")
         )
-        content = content.strip()
-        content = re.sub(r"^```(?:json)?\s*|\s*```$", "", content, flags=re.IGNORECASE | re.DOTALL).strip()
-        data = json.loads(content)
+        data = self.parse_json_object(content)
         if not isinstance(data, dict):
             raise ModelError("compactor returned non-object JSON")
         return data
 
+    @classmethod
+    def parse_json_object(cls, text: str) -> Json:
+        text = cls.strip_json_fence(Text.clean(text).strip())
+        if not text:
+            raise ModelError("compactor returned empty output")
+        try:
+            data = json.loads(text)
+        except json.JSONDecodeError:
+            data = repair_json(text, return_objects=True)
+        if isinstance(data, dict):
+            return data
+        raise ModelError("compactor returned invalid JSON: " + Tool.compact(text, 200))
+
+    @staticmethod
+    def strip_json_fence(text: str) -> str:
+        match = re.match(r"^```(?:json)?\s*(.*?)\s*```$", text, flags=re.IGNORECASE | re.DOTALL)
+        return (match.group(1) if match else text).strip()
+
     def client(self) -> OpenAI:
         provider = self.session.config.provider
         if missing := self.session.missing_config():
@@ -3238,13 +3373,13 @@ FLOW:
 - ACT when clear; keep using tools until done.
 - Every turn must call tools or return final; never emit empty content.
 - Prefer built-ins over Bash. Batch independent read-only calls.
-- All user-visible output follows the latest user input and uses the user's language.
-- Interim text is shown to the user and kept as conversation.
+- All assistant text is user-visible, including interim narration before tool calls and final answers.
+- Write every assistant text message as markdown in the latest user's language; do not switch languages unless asked.
 
 CONTEXT:
 - Trust LATEST FILE STATE from Read/Edit for listed ranges.
 - Recall bounded tr.N only when needed; prefer FILE STATE over old outputs.
-- For multi-step work, call Note early with goal and a short plan; update plan/known when they change.
+- For multi-step work, call Note early with goal and a short plan; update plan/known/check when they change.
 
 EDITS:
 - Inspect/read before edits.
@@ -3256,7 +3391,8 @@ EDITS:
 
 FINAL:
 - Concise markdown in the user's language.
-- Include changed files and checks when relevant.\
+- Include changed files and checks when relevant.
+- Mention checks run, or say checks not run.\
 """
 
     def __init__(self, session: Session, input_fn=input, output_fn=print):
@@ -3445,7 +3581,7 @@ class UiPrinter:
             return [("ansicyan", text + "\n")]
         if text.startswith("Error:") or text.startswith("ConfigError:") or text.startswith("Unknown command:"):
             return [("ansired", text + "\n")]
-        return self.text_segments(text)
+        return [("ansiwhite", line + "\n") for line in text.splitlines() or [""]]
 
     def tool_segments(self, text: str) -> list[tuple[str, str]]:
         segments = []
@@ -3562,10 +3698,6 @@ class UiPrinter:
                 at_start = part.endswith("\n")
         return indented
 
-    def text_segments(self, text: str) -> list[tuple[str, str]]:
-        return [("ansiwhite", line + "\n") for line in text.splitlines() or [""]]
-
-
 class BashLivePreview:
     HEIGHT: ClassVar[int] = 6
     MAX_CHARS: ClassVar[int] = 8000
@@ -3737,15 +3869,13 @@ class StatusBar:
             self.output.flush()
             self.rendered = False
 
-    def elapsed(self) -> float:
-        return max(0.0, time.monotonic() - self.started_at) if self.started_at else 0.0
-
     def idle_fragments(self) -> list[tuple[str, str]]:
         return self.fragments(0.0, sweep=False, show_elapsed=False)
 
     def active_fragments(self) -> list[tuple[str, str]]:
         self.refresh_retry_state()
-        return self.fragments(self.elapsed(), sweep=True, show_elapsed=True)
+        elapsed = max(0.0, time.monotonic() - self.started_at) if self.started_at else 0.0
+        return self.fragments(elapsed, sweep=True, show_elapsed=True)
 
     def fragments(self, elapsed: float, *, sweep: bool, show_elapsed: bool) -> list[tuple[str, str]]:
         text = self.text(elapsed, show_elapsed=show_elapsed)
@@ -3870,6 +4000,7 @@ Tools:
         self.live_status_paused = False
         self.live_queue_paused = False
         self.transient_tool_lines = 0
+        self.approval_full_preview = ""
         self.interactive_input = input_fn is input and sys.stdin.isatty()
         self.queue_input_paused = threading.Event()
         self.queue_input_active = threading.Event()
@@ -3888,6 +4019,7 @@ Tools:
         self.agent.tools.output_fn = self.tool_output
         self.agent.tools.input_fn = self.tool_input
         self.agent.tools.preview_fn = self.tool_preview
+        self.agent.tools.preview_full_fn = lambda text: setattr(self, "approval_full_preview", text)
         self.agent.tools.live_start = self.tool_live_start
         self.agent.tools.live_output = self.tool_live_output
 
@@ -4094,7 +4226,14 @@ Tools:
         frame = "|/-\\"[int(time.monotonic() / 0.2) % 4]
         return [("class:approval", prompt_text), ("class:approval.wait", frame + " ")]
 
-    def read_input(self, prompt_text: str = "nano> ", *, multiline: bool = False, submit_on_enter: bool = False, prompt_style: str = "class:prompt") -> str:
+    def read_input(
+        self,
+        prompt_text: str = "nano> ",
+        *,
+        multiline: bool = False,
+        submit_on_enter: bool = False,
+        prompt_style: str = "class:prompt",
+    ) -> str:
         if self.input_history is None:
             return self.input_fn(prompt_text)
 
@@ -4150,6 +4289,10 @@ Tools:
             else:
                 pt_search.start_search(direction=direction)
 
+        @bindings.add("c-a", filter=Condition(lambda: bool(self.approval_full_preview)), eager=True)
+        def _ctrl_a(event):
+            run_in_terminal(self.open_approval_preview)
+
         @bindings.add("tab")
         def _tab(event):
             if buffer.complete_state:
@@ -4209,7 +4352,12 @@ Tools:
         if text.startswith("approve ") and self.interactive_input and sys.stdout.isatty():
             self.with_status_paused(lambda: self.show_transient_tool_output(text))
             return
-        self.with_status_paused(lambda: self.emit_tool_output(text))
+
+        def emit() -> None:
+            self.clear_transient_tool_output()
+            self.emit(text)
+
+        self.with_status_paused(emit)
 
     def agent_output(self, text: str = "") -> None:
         self.with_status_paused(lambda: self.emit_agent_output(text))
@@ -4225,6 +4373,7 @@ Tools:
             finally:
                 if self.interactive_input and sys.stdout.isatty():
                     self.clear_transient_tool_output()
+                    self.approval_full_preview = ""
 
         return self.with_status_paused(read)
 
@@ -4239,20 +4388,46 @@ Tools:
         self.with_status_paused(lambda: self.show_transient_tool_preview(text))
         return True
 
+    def open_approval_preview(self) -> None:
+        if not self.approval_full_preview:
+            return
+        fd, path = tempfile.mkstemp(prefix="nanocode-preview-", suffix=".diff")
+        try:
+            pager_env = os.environ.get("PAGER", "")
+            pager = shlex.split(pager_env) if pager_env else ([less, "-R"] if (less := shutil.which("less")) else [])
+            text = self.ansi_diff_preview(self.approval_full_preview) if pager and os.path.basename(pager[0]) == "less" else self.approval_full_preview
+            with os.fdopen(fd, "w", encoding="utf-8") as file:
+                file.write(text.rstrip() + "\n")
+            if pager:
+                subprocess.run([*pager, path])
+            else:
+                print(self.approval_full_preview)
+                input("Press Enter to return...")
+        finally:
+            try:
+                os.unlink(path)
+            except OSError:
+                pass
+
+    @staticmethod
+    def ansi_diff_preview(text: str) -> str:
+        colors = [("---", "\033[90m"), ("+++", "\033[90m"), ("@@", "\033[36m"), ("+", "\033[32m"), ("-", "\033[31m")]
+        lines = []
+        for line in text.splitlines():
+            style = next((color for prefix, color in colors if line.lstrip().startswith(prefix)), "")
+            lines.append(style + line + ("\033[0m" if style else ""))
+        return "\n".join(lines)
+
     def show_transient_tool_preview(self, text: str) -> None:
         self.clear_transient_tool_output()
         lines = text.rstrip().splitlines()
         if not lines:
             return
         height, width = 12, max(20, shutil.get_terminal_size((120, 20)).columns)
-        shown = lines[:height] + (["... preview truncated ..."] if len(lines) > height else [])
+        shown = lines[:height] + ([f"... preview truncated: {len(lines) - height} more lines (Ctrl-A: full preview) ..."] if len(lines) > height else [])
         self.emit("\n".join(line[: max(0, width - 1)] for line in shown))
         self.transient_tool_lines = len(shown)
 
-    def emit_tool_output(self, text: str) -> None:
-        self.clear_transient_tool_output()
-        self.emit(text)
-
     def emit_agent_output(self, text: str) -> None:
         self.clear_transient_tool_output()
         if self.ui.color and text.strip():
@@ -4651,16 +4826,20 @@ Tools:
         compacted, keep = self.agent.context.compaction_parts()
         if not compacted:
             return "No prior conversation to compact"
+        fallback = False
         try:
             self.status_bar.start()
             data = self.agent.model.compact(self.agent.context.compaction_input(compacted))
         except KeyboardInterrupt:
             return "Cancelled"
-        except Exception as error:
-            return "Error: " + str(error)
+        except Exception:
+            self.agent.context.apply_compaction_fallback(keep)
+            fallback = True
+            data = None
         finally:
             self.status_bar.stop()
-        self.agent.context.apply_compaction(data, keep)
+        if data is not None:
+            self.agent.context.apply_compaction(data, keep)
         self.agent.context.update_percent(self.agent.context.model_messages(self.agent.SYSTEM_PROMPT))
         return (
             "Compacted context: messages "
@@ -4670,6 +4849,7 @@ Tools:
             + ", prior summary inserted, ctx "
             + str(self.session.state.context_percent)
             + "%"
+            + (" (fallback)" if fallback else "")
         )
 
     def index(self, args: str) -> str:

{nanocode_cli-0.5.7 → nanocode_cli-0.5.10/nanocode_cli.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nanocode-cli
-Version: 0.5.7
+Version: 0.5.10
 Summary: A small terminal coding agent written in Python
 Author-email: hit9 <hit9@icloud.com>
 License-Expression: BSD-3-Clause
@@ -23,6 +23,7 @@ Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: anthropic>=0.64.0
 Requires-Dist: code-symbol-index>=0.1.13
+Requires-Dist: json-repair
 Requires-Dist: openai>=2.37.0
 Requires-Dist: prompt-toolkit>=3.0
 Requires-Dist: rich>=13.0
@@ -36,19 +37,22 @@ Dynamic: license-file
 
 A small terminal coding agent written in Python.
 
+[简体中文](README.zh-CN.md)
+
 nanocode is pre-1.0 software. Commands, configuration, and tool behavior may change before a stable release.
 
 ![nanocode screenshot](snapshots/nanocode-snapshot.png)
 
 ## Features
 
-- **Current turn flow**: Interim answers, tool results, and appended user input stay in order during a running task.
-- **Latest file state**: `Read` and `Edit` maintain a current, line-numbered file view with stale-range checks.
-- **Anchored edits**: `line:hash` anchors catch stale edits before they touch the wrong code.
-- **Working memory**: `Note` keeps the active goal, plan, and known facts separate from noisy tool output.
-- **Symbol index**: Jump from names to outlines, references, and changed files without searching blindly.
-- **Tool recall**: Prompt output is bounded, while raw `tr.N` results remain recallable when needed.
-- **Terminal-native UI**: Model picking, history search, confirmations, live command output, appended input, and status stay in the terminal.
+- **Live turn control**: Add follow-up input while the agent is still working, without losing the current tool flow.
+- **File-state brain**: Reads and edits build a current, line-numbered view of the files that matter now.
+- **Stale-edit protection**: `line:hash` anchors reject edits when the target code has drifted.
+- **Project-aware navigation**: Use the symbol index to jump through outlines, references, and changed files quickly.
+- **Recoverable context**: Tool output stays bounded in the prompt, while raw `tr.N` results remain recallable.
+- **Cache-aware context**: Stable sections stay early and noisy working state stays late to improve prompt-cache reuse.
+- **Focused working memory**: `Note` separates goal, plan, and known facts from noisy execution logs.
+- **Terminal-first workflow**: Model selection, history search, confirmations, live command output, appended input, and status all stay in one CLI.
 
 ## Install
 
@@ -56,6 +60,12 @@ nanocode is pre-1.0 software. Commands, configuration, and tool behavior may cha
 uv tool install nanocode-cli
 ```
 
+Upgrade:
+
+```sh
+uv tool upgrade nanocode-cli
+```
+
 For local development:
 
 ```sh

{nanocode_cli-0.5.7 → nanocode_cli-0.5.10}/nanocode_cli.egg-info/SOURCES.txt

@@ -1,6 +1,7 @@
 LICENSE
 MANIFEST.in
 README.md
+README.zh-CN.md
 nanocode.py
 pyproject.toml
 nanocode_cli.egg-info/PKG-INFO

{nanocode_cli-0.5.7 → nanocode_cli-0.5.10}/nanocode_cli.egg-info/requires.txt

@@ -1,5 +1,6 @@
 anthropic>=0.64.0
 code-symbol-index>=0.1.13
+json-repair
 openai>=2.37.0
 prompt-toolkit>=3.0
 rich>=13.0

{nanocode_cli-0.5.7 → nanocode_cli-0.5.10}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "nanocode-cli"
-version = "0.5.7"
+version = "0.5.10"
 description = "A small terminal coding agent written in Python"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -29,6 +29,7 @@ classifiers = [
 dependencies = [
     "anthropic>=0.64.0",
     "code-symbol-index>=0.1.13",
+    "json-repair",
     "openai>=2.37.0",
     "prompt-toolkit>=3.0",
     "rich>=13.0",