auto-coder 3.0.71__tar.gz → 3.0.72__tar.gz

{auto_coder-3.0.71/src/auto_coder.egg-info → auto_coder-3.0.72}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: auto-coder
-Version: 3.0.71
+Version: 3.0.72
 Summary: AutoCoder: AutoCoder
 Author: allwefantasy
 Author-email: allwefantasy <allwefantasy@gmail.com>

{auto_coder-3.0.71 → auto_coder-3.0.72}/pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "auto-coder"
-version = "3.0.71"
+version = "3.0.72"
 description = "AutoCoder: AutoCoder"
 authors = [
     {name = "allwefantasy", email = "allwefantasy@gmail.com"}

{auto_coder-3.0.71 → auto_coder-3.0.72/src/auto_coder.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: auto-coder
-Version: 3.0.71
+Version: 3.0.72
 Summary: AutoCoder: AutoCoder
 Author: allwefantasy
 Author-email: allwefantasy <allwefantasy@gmail.com>

{auto_coder-3.0.71 → auto_coder-3.0.72}/src/auto_coder.egg-info/SOURCES.txt

@@ -1056,6 +1056,7 @@ src/autocoder/terminal_core/forward_input.py
 src/autocoder/terminal_core/index_rules.py
 src/autocoder/terminal_core/lifecycle.py
 src/autocoder/terminal_core/output.py
+src/autocoder/terminal_core/session_rag_search.py
 src/autocoder/terminal_core/status.py
 src/autocoder/terminal_core/task_runtime.py
 src/autocoder/terminal_core/tunnel_worker.py
@@ -1209,6 +1210,7 @@ tests/test_sdk_response_models.py
 tests/test_sdk_task_completion_provider.py
 tests/test_sdk_validators.py
 tests/test_semantic_index_refresh.py
+tests/test_session_rag_search.py
 tests/test_shell_bang_command_result_rendering.py
 tests/test_shell_command_robustness.py
 tests/test_stream_chat_with_continue.py

{auto_coder-3.0.71 → auto_coder-3.0.72}/src/autocoder/auto_coder_terminal_v3.py

@@ -70,6 +70,7 @@ V3_COMMANDS = [
     "/refresh",
     "/rules",
     "/service",
+    "/session",
     "/sessions",
     "/side",
     "/status",

{auto_coder-3.0.71 → auto_coder-3.0.72}/src/autocoder/command_skills.py

@@ -48,6 +48,8 @@ Common scenarios:
   auto-coder.chat.lite --no-mouse-support
   Inside the TUI: /status (agentic phase + tool-call implementation; effective profile; /bg flash; cache/budget input; subagents + worker pair)
   Inside the TUI: /sessions /list (latest 50; title 28 chars; summary 120 chars)
+  Inside the TUI: /session /query backend transcript search
+  Without RAG model config: /session /query uses the local session cache
   Inside the TUI: /sessions /path
   Inside the TUI: /bg -1 (resume the last background session)
   Inside the TUI: /bg parser (parallel title/summary/conversations match)
@@ -76,6 +78,7 @@ Common scenarios:
   Inside the TUI: /enable_subagents /disable
   Inside the TUI: /enable_subagents /reset
   Inside the TUI: /conf enable_kimi_optimize:true
+  Inside the TUI: /conf rag_model:infinisynapse/deepseek/deepseek-v4-flash
   Inside the TUI: /conf agent_input_token_budget:300k
   Inside the TUI: /conf agent_output_token_budget:32k
   Inside the TUI: /conf agent_input_cost_budget:1.5
@@ -786,6 +789,7 @@ Inside the TUI, use slash commands:
 /fusion --fusion_code_models codex:gpt-5.5:2 --fusion_review_models cursor::1 fix the failing tests
 /new
 /sessions /list
+/session /query backend transcript search
 /sessions /path
 /bg
 /bg -1
@@ -953,6 +957,17 @@ Each row's `Summary:` contains the latest user-initiated task or question;
 TUI control commands and internal agent/tool messages are excluded. Titles are
 limited to 28 characters and summaries to 120 characters; truncated text ends
 with `...`.
+`/session /query <keyword>` (also `/sessions /query`) builds or incrementally
+refreshes `.auto-coder/session-rag-cache/` from all current-project canonical
+Auto-Coder conversations plus their bound backend transcripts, then invokes
+`auto-coder.rag` with explicit recall/chunk/QA model arguments. Configure one
+shared alias with `/conf rag_model:<alias>`, or configure the three stages with
+`/conf rag_recall_model:<alias>`, `/conf rag_chunk_model:<alias>`, and
+`/conf rag_qa_model:<alias>`. RAG output is treated as untrusted: only existing
+current-project Auto-Coder session IDs are shown. If neither `rag_model` nor a
+complete three-stage profile is configured, the command skips the model call
+and performs a local case-insensitive fixed-string search over the same cache;
+whitespace-separated terms must all occur, with exact phrases ranked first.
 `/sessions /path` prints the foreground Auto-Coder canonical conversation JSON
 path. `/side /sessions /path` prints the active side conversation's Auto-Coder
 canonical JSON path. These commands never return Codex/Cursor native session

{auto_coder-3.0.71 → auto_coder-3.0.72}/src/autocoder/common/__init__.py

@@ -254,6 +254,7 @@ class AutoCoderArgs(pydantic.BaseModel):
     execute: Optional[bool] = None
 
     model: str = ""
+    rag_model: str = ""
     rag_recall_model: str = ""
     rag_chunk_model: str = ""
     rag_qa_model: str = ""

{auto_coder-3.0.71 → auto_coder-3.0.72}/src/autocoder/common/conf_validator.py

@@ -78,6 +78,11 @@ class ConfigValidator:
             "description": "产品模式(lite/pro)",
         },
         "model": {"type": str, "default": "v3_chat", "description": "默认模型名称"},
+        "rag_model": {
+            "type": str,
+            "default": "",
+            "description": "RAG 召回、分块与问答共用的模型 alias",
+        },
         "rag_recall_model": {
             "type": str,
             "default": "",
@@ -432,6 +437,7 @@ class ConfigValidator:
             "voice2text_model",
             "commit_model",
             "model",
+            "rag_model",
             "rag_recall_model",
             "rag_chunk_model",
             "rag_qa_model",

{auto_coder-3.0.71 → auto_coder-3.0.72}/src/autocoder/common/test_autocoder_args.py

@@ -21,6 +21,7 @@ class TestAutoCoderArgsDefaults:
         
         # Test model fields
         assert args.model == ""
+        assert args.rag_model == ""
         assert args.agent_backend == ""
         assert args.agent_backend_model == ""
         assert args.codex_model == ""

{auto_coder-3.0.71 → auto_coder-3.0.72}/src/autocoder/common/v2/agent/agentic_edit_tools/extra_docs/autocoder_cli_reference.py

@@ -589,6 +589,7 @@ def _autocoder_cli_reference() -> str:
     ```text
     /sessions
     /sessions /list
+    /session /query <keyword>
     /sessions /current
     /sessions /path
     /sessions /resume <id|name|#index>
@@ -603,6 +604,13 @@ def _autocoder_cli_reference() -> str:
     question. TUI control commands and internal agent/tool messages do not
     replace that summary. Titles are limited to 28 characters and summaries to
     120 characters; truncated text ends with `...`.
+    `/session /query` (also `/sessions /query`) indexes canonical conversations
+    and their bound backend transcripts under `.auto-coder/session-rag-cache/`,
+    calls `auto-coder.rag`, and filters its returned IDs against actual project
+    sessions. Configure it with `rag_model` or all of `rag_recall_model`,
+    `rag_chunk_model`, and `rag_qa_model` through `/conf`. With no complete RAG
+    profile it skips `auto-coder.rag` and uses a local case-insensitive keyword
+    search over those cached documents instead.
     `/sessions /path` prints only the current foreground conversation's
     Auto-Coder canonical JSON path. It does not print a Codex/Cursor native
     session path.

{auto_coder-3.0.71 → auto_coder-3.0.72}/src/autocoder/completer/command_completer_v2.py

@@ -229,6 +229,16 @@ COMMAND_HIERARCHY = {
     },
     "/sessions": {
         "/list": {},
+        "/query": {},
+        "/current": {},
+        "/path": {},
+        "/resume": {},
+        "/rename": {},
+        "/help": {},
+    },
+    "/session": {
+        "/list": {},
+        "/query": {},
         "/current": {},
         "/path": {},
         "/resume": {},
@@ -423,6 +433,7 @@ class CommandCompleterV2(Completer):
             "/summon": self._handle_text_with_symbols,
             "/design": self._handle_design,
             "/rules": self._handle_rules,
+            "/session": self._handle_sessions,
             "/sessions": self._handle_sessions,
             "/workflow": self._handle_workflow,
             "/rags": self._handle_rags,
@@ -1468,13 +1479,14 @@ class CommandCompleterV2(Completer):
         word: str,
         text: str,
     ) -> Iterable[Completion]:
-        """Handles completions for top-level /sessions command."""
-        args_text = text[len("/sessions") :].lstrip()
+        """Handles completions for top-level /session(s) commands."""
+        command_prefix = "/sessions" if text.startswith("/sessions") else "/session"
+        args_text = text[len(command_prefix) :].lstrip()
         parts = args_text.split()
         last_part = parts[-1] if parts and not text.endswith(" ") else ""
 
         if not args_text or (len(parts) == 1 and not text.endswith(" ")):
-            for sub_cmd in COMMAND_HIERARCHY["/sessions"]:
+            for sub_cmd in COMMAND_HIERARCHY[command_prefix]:
                 if sub_cmd.startswith(last_part):
                     yield Completion(sub_cmd, start_position=-len(last_part))
             return

{auto_coder-3.0.71 → auto_coder-3.0.72}/src/autocoder/rag/model_profile.py

@@ -18,6 +18,7 @@ from typing import Any
 from filelock import FileLock
 
 RAG_MODEL_PROFILE_SCHEMA_VERSION = 1
+RAG_SINGLE_MODEL_CONFIG_KEY = "rag_model"
 RAG_STAGE_CONFIG_KEYS = {
     "recall_model": "rag_recall_model",
     "chunk_model": "rag_chunk_model",
@@ -115,6 +116,31 @@ def _scoped_value(
     return "", ""
 
 
+def _scoped_stage_value(
+    key: str,
+    *,
+    overrides: Mapping[str, str],
+    project_conf: Mapping[str, Any],
+    global_conf: Mapping[str, Any],
+    allow_rag_model_fallback: bool,
+) -> tuple[str, str]:
+    """Resolve one stage while preserving configuration-scope precedence."""
+
+    for scope, values in (
+        ("override", overrides),
+        ("project", project_conf),
+        ("global", global_conf),
+    ):
+        stage_value = _nonempty(values.get(key))
+        if stage_value:
+            return stage_value, f"{scope}:{key}"
+        if allow_rag_model_fallback:
+            shared_value = _nonempty(values.get(RAG_SINGLE_MODEL_CONFIG_KEY))
+            if shared_value:
+                return shared_value, f"{scope}:{RAG_SINGLE_MODEL_CONFIG_KEY}"
+    return "", ""
+
+
 def _profile_fingerprint(payload: Mapping[str, Any]) -> str:
     encoded = json.dumps(
         payload,
@@ -131,12 +157,15 @@ def resolve_rag_model_profile(
     overrides: Mapping[str, str] | None = None,
     product_mode: str | None = None,
     model_file: str | None = None,
+    allow_rag_model_fallback: bool = False,
 ) -> RAGModelProfile:
     """Resolve and validate one run-scoped RAG model profile.
 
     Project configuration wins over global configuration. All three stage
     aliases are explicit opt-in settings: the general ``model`` and legacy
     ``v3_chat`` alias never enable an automatic RAG consumer implicitly.
+    Callers that explicitly support a shared ``rag_model`` alias may opt into
+    that fallback without changing existing three-stage consumers.
     """
 
     overrides = dict(overrides or {})
@@ -187,11 +216,12 @@ def resolve_rag_model_profile(
     stage_values: dict[str, str] = {}
     sources: dict[str, str] = {}
     for stage, config_key in RAG_STAGE_CONFIG_KEYS.items():
-        value, source = _scoped_value(
+        value, source = _scoped_stage_value(
             config_key,
             overrides=overrides,
             project_conf=project_conf,
             global_conf=global_conf,
+            allow_rag_model_fallback=allow_rag_model_fallback,
         )
         stage_values[stage] = value
         sources[stage] = source
@@ -239,12 +269,17 @@ def resolve_rag_model_profile(
                     bool(model is not None and model.has_api_key),
                 )
             exists, has_credential = checked[alias]
+            diagnostic_field = (
+                sources[stage].split(":", 1)[1]
+                if ":" in sources[stage]
+                else RAG_STAGE_CONFIG_KEYS[stage]
+            )
             if not exists:
                 invalid = True
                 diagnostics.append(
                     {
                         "code": "model_not_found",
-                        "field": RAG_STAGE_CONFIG_KEYS[stage],
+                        "field": diagnostic_field,
                         "message": f"Configured model alias does not exist: {alias}",
                     }
                 )
@@ -253,7 +288,7 @@ def resolve_rag_model_profile(
                 diagnostics.append(
                     {
                         "code": "model_credential_unavailable",
-                        "field": RAG_STAGE_CONFIG_KEYS[stage],
+                        "field": diagnostic_field,
                         "message": f"Configured model alias has no usable credential: {alias}",
                     }
                 )
@@ -309,6 +344,7 @@ def resolve_rag_model_profile(
 __all__ = [
     "RAGModelProfile",
     "RAG_MODEL_PROFILE_SCHEMA_VERSION",
+    "RAG_SINGLE_MODEL_CONFIG_KEY",
     "RAG_STAGE_CONFIG_KEYS",
     "resolve_rag_model_profile",
 ]

{auto_coder-3.0.71 → auto_coder-3.0.72}/src/autocoder/sdk/cli/help_docs.py

@@ -40,6 +40,7 @@ Description:
 
 Examples:
   auto-coder.run config model=infinisynapse/deepseek/deepseek-v4-flash
+  auto-coder.run config rag_model=infinisynapse/deepseek/deepseek-v4-flash
   auto-coder.run config code_model=volcengine/coding/doubao-seed-2.0-pro
   auto-coder.run config vl_model=winclaw/vl
   auto-coder.run config enable_kimi_optimize=true
@@ -50,6 +51,11 @@ Examples:
 
 Configuration keys:
   model                          Default primary model
+  rag_model                      Optional shared recall/chunk/QA model for
+                                 session search; absent uses local cache grep
+  rag_recall_model               RAG document recall/relevance model
+  rag_chunk_model                RAG long-document chunk extraction model
+  rag_qa_model                   RAG final-answer model
   code_model                     Model used for code generation
   vl_model                       Vision-language model
   enable_kimi_optimize           true/false — opt-in Kimi-style loop optimization:
@@ -151,6 +157,9 @@ the same `configure()` function used by `/conf` in the interactive terminal.
 # Default primary model
 auto-coder.run config model=infinisynapse/deepseek/deepseek-v4-flash
 
+# One shared model for /session /query (or set all three stage-specific keys)
+auto-coder.run config rag_model=infinisynapse/deepseek/deepseek-v4-flash
+
 # Dedicated models for code and vision tasks
 auto-coder.run config code_model=volcengine/coding/doubao-seed-2.0-pro vl_model=winclaw/vl
 
@@ -175,6 +184,10 @@ auto-coder.run config subagent_backend=codex subagent_model=default orchestratio
 | Key | Type | Description |
 |-----|------|-------------|
 | model | model name | Default primary model for the project |
+| rag_model | model name | Optional shared recall, chunk, and QA model for `/session /query`; without a complete RAG profile the command searches its local cache |
+| rag_recall_model | model name | Recall and document-relevance model; overrides `rag_model` for this stage |
+| rag_chunk_model | model name | Long-document chunk-extraction model; overrides `rag_model` for this stage |
+| rag_qa_model | model name | Final-answer model; overrides `rag_model` for this stage |
 | code_model | model name | Model used for code generation |
 | vl_model | model name | Vision-language model for image input |
 | enable_kimi_optimize | true / false | Opt-in Kimi-style loop optimization. Plain assistant text can complete the turn; empty or malformed responses get up to 10 attempts (aligned with kimi-code); identical consecutive tool calls receive escalating reminders and are force-stopped at 12. Default: `false`. |
@@ -336,6 +349,7 @@ Persist per-project model and orchestration settings:
 
 ```bash
 auto-coder.run config model=infinisynapse/deepseek/deepseek-v4-flash
+auto-coder.run config rag_model=infinisynapse/deepseek/deepseek-v4-flash
 auto-coder.run config code_model=volcengine/coding/doubao-seed-2.0-pro
 auto-coder.run config enable_subagent_orchestration=true
 auto-coder.run config subagent_backend=inherit
@@ -348,7 +362,8 @@ auto-coder.run config goal-verifier-model=gpt-5.5
 auto-coder.run models /list    # discover valid model names
 ```
 
-Keys: `model`, `code_model`, `vl_model`, `enable_subagent_orchestration`,
+Keys: `model`, `rag_model`, `rag_recall_model`, `rag_chunk_model`,
+`rag_qa_model`, `code_model`, `vl_model`, `enable_subagent_orchestration`,
 `subagent_backend`, `subagent_model`,
 `orchestration_worker_isolation`, `orchestration_max_concurrency`,
 `orchestration_max_depth`,

{auto_coder-3.0.71 → auto_coder-3.0.72}/src/autocoder/terminal_core/backends.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import copy
 import hashlib
 import json
 import os
@@ -438,6 +439,32 @@ class BackendSessionStore:
         metadata = binding.get("metadata")
         return metadata if isinstance(metadata, dict) else {}
 
+    def list_bindings(self) -> dict[str, dict[str, dict[str, Any]]]:
+        """Return a detached snapshot of every canonical-to-native binding."""
+
+        with self._file_lock():
+            data = self._load_unlocked()
+        result: dict[str, dict[str, dict[str, Any]]] = {}
+        conversations = data.get("conversations")
+        if not isinstance(conversations, dict):
+            return result
+        for conversation_id, conversation in conversations.items():
+            if not isinstance(conversation, dict):
+                continue
+            bindings = conversation.get("bindings")
+            if not isinstance(bindings, dict):
+                continue
+            valid = {
+                str(backend): copy.deepcopy(binding)
+                for backend, binding in bindings.items()
+                if isinstance(binding, dict)
+                and isinstance(binding.get("session_id"), str)
+                and binding.get("session_id")
+            }
+            if valid:
+                result[str(conversation_id)] = valid
+        return result
+
     def set_session_id(
         self,
         conversation_id: str,