klaude-code 2.19.0__tar.gz → 2.21.0__tar.gz

{klaude_code-2.19.0 → klaude_code-2.21.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: klaude-code
-Version: 2.19.0
+Version: 2.21.0
 Summary: Minimal code agent CLI
 Requires-Dist: anthropic>=0.66.0
 Requires-Dist: chardet>=5.2.0,<6
@@ -29,7 +29,7 @@ Minimal code agent CLI.
 - **Keep reasoning item in context**: Interleaved thinking support
 - **Model-aware tools**: Claude Code tool set for Opus, `apply_patch` for GPT-5/Codex
 - **Reminders**: Cooldown-based todo tracking, instruction reinforcement and external file change reminder
-- **Sub-agents**: Task, Explore, Web, ImageGen
+- **Sub-agents**: Task, Explore, Web
 - **Structured sub-agent output**: Main agent defines JSON schema and get schema-compliant responses via constrained decoding
 - **Recursive `@file` mentions**: Circular dependency protection, relative path resolution
 - **External file sync**: Monitoring for external edits (linter, manual)
@@ -37,7 +37,6 @@ Minimal code agent CLI.
 - **Output truncation**: Large outputs saved to file system with snapshot links
 - **Agent Skills**: Built-in + user + project Agent Skills (with implicit invocation by Skill tool or explicit invocation by typing `//skill` or `/skill`)
 - **Sessions**: Resumable with `--continue`
-- **Mermaid diagrams**: Terminal image preview and Interactive local HTML viewer with zoom, pan, and SVG export
 - **Extras**: Slash commands, sub-agents, image paste, terminal notifications, auto-theming
 
 ## Installation
@@ -275,4 +274,3 @@ The main agent can spawn specialized sub-agents for specific tasks:
 | **Explore** | Fast codebase exploration - find files, search code, answer questions about the codebase |
 | **Task** | Handle complex multi-step tasks autonomously |
 | **WebAgent** | Search the web, fetch pages, and analyze content. Uses Brave LLM Context API when `BRAVE_API_KEY` is set, otherwise falls back to DuckDuckGo |
-| **ImageGen** | Generate images from text prompts via OpenRouter Nano Banana Pro |

{klaude_code-2.19.0 → klaude_code-2.21.0}/README.md

@@ -7,7 +7,7 @@ Minimal code agent CLI.
 - **Keep reasoning item in context**: Interleaved thinking support
 - **Model-aware tools**: Claude Code tool set for Opus, `apply_patch` for GPT-5/Codex
 - **Reminders**: Cooldown-based todo tracking, instruction reinforcement and external file change reminder
-- **Sub-agents**: Task, Explore, Web, ImageGen
+- **Sub-agents**: Task, Explore, Web
 - **Structured sub-agent output**: Main agent defines JSON schema and get schema-compliant responses via constrained decoding
 - **Recursive `@file` mentions**: Circular dependency protection, relative path resolution
 - **External file sync**: Monitoring for external edits (linter, manual)
@@ -15,7 +15,6 @@ Minimal code agent CLI.
 - **Output truncation**: Large outputs saved to file system with snapshot links
 - **Agent Skills**: Built-in + user + project Agent Skills (with implicit invocation by Skill tool or explicit invocation by typing `//skill` or `/skill`)
 - **Sessions**: Resumable with `--continue`
-- **Mermaid diagrams**: Terminal image preview and Interactive local HTML viewer with zoom, pan, and SVG export
 - **Extras**: Slash commands, sub-agents, image paste, terminal notifications, auto-theming
 
 ## Installation
@@ -253,4 +252,3 @@ The main agent can spawn specialized sub-agents for specific tasks:
 | **Explore** | Fast codebase exploration - find files, search code, answer questions about the codebase |
 | **Task** | Handle complex multi-step tasks autonomously |
 | **WebAgent** | Search the web, fetch pages, and analyze content. Uses Brave LLM Context API when `BRAVE_API_KEY` is set, otherwise falls back to DuckDuckGo |
-| **ImageGen** | Generate images from text prompts via OpenRouter Nano Banana Pro |

{klaude_code-2.19.0 → klaude_code-2.21.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "uv_build"
 
 [project]
 name = "klaude-code"
-version = "2.19.0"
+version = "2.21.0"
 description = "Minimal code agent CLI"
 readme = "README.md"
 requires-python = ">=3.13"
@@ -40,6 +40,7 @@ dev = [
     "pyright>=1.1.407",
     "pytest>=8.4.1",
     "pytest-cov>=7.0.0",
+    "ruff==0.14.3",
     "ty>=0.0.8",
 ]
 
@@ -69,7 +70,10 @@ venv = ".venv"
 extraPaths = ["src"]
 reportMissingImports = "warning"
 reportMissingModuleSource = "warning"
-exclude = [".venv/"]
+exclude = [
+    ".venv/",
+    "src/klaude_code/skill/assets/**/scripts/",
+]
 
 [[tool.pyright.executionEnvironments]]
 root = "."

{klaude_code-2.19.0 → klaude_code-2.21.0}/src/klaude_code/cli/log_viewer.html

@@ -93,10 +93,13 @@ body {
   flex-wrap: wrap;
 }
 .filter-btn {
-  background: transparent;
-  border: 1px solid transparent;
+  display: inline-flex;
+  align-items: center;
+  gap: 6px;
+  background: var(--bg-surface);
+  border: 1px solid var(--border-strong);
   color: var(--fg-secondary);
-  padding: 4px 8px;
+  padding: 4px 10px;
   border-radius: 6px;
   font-size: 16px;
   font-weight: 700;
@@ -105,13 +108,23 @@ body {
   transition: all 0.2s ease;
   line-height: 1.2;
 }
-.filter-btn:hover { background: #e6f0fa; color: #005fcc; }
+.filter-btn:hover {
+  background: var(--bg-hover);
+  color: var(--fg);
+  border-color: var(--fg-tertiary);
+}
 .filter-btn.active { 
   background: #dbeafe; 
   color: #1e3a8a; 
-  border-color: transparent; 
+  border-color: #bfdbfe;
   box-shadow: none; 
 }
+.filter-btn.active::after {
+  content: '\00d7';
+  font-size: 14px;
+  line-height: 1;
+  opacity: 0.85;
+}
 
 #search-box {
   background: var(--bg-surface);

{klaude_code-2.19.0 → klaude_code-2.21.0}/src/klaude_code/cli/main.py

@@ -35,7 +35,6 @@ def _build_env_help() -> str:
             "  KLAUDE_READ_GLOBAL_LINE_CAP    Max lines to read (default: 2000)",
             "  KLAUDE_READ_MAX_CHARS          Max total chars to read (default: 50000)",
             "  KLAUDE_READ_MAX_IMAGE_BYTES    Max image bytes to read (default: 64MB)",
-            "  KLAUDE_IMAGE_OUTPUT_MAX_BYTES  Max decoded image bytes (default: 64MB)",
         ]
     )
     return "\n\n".join(lines)
@@ -188,12 +187,6 @@ def main_callback(
         "--vanilla",
         help="Minimal mode: basic tools only, no system prompts",
     ),
-    banana: bool = typer.Option(
-        False,
-        "--banana",
-        help="Image generation mode (alias for --model banana)",
-        rich_help_panel="LLM",
-    ),
     version: bool = typer.Option(
         False,
         "--version",
@@ -208,10 +201,6 @@ def main_callback(
     if ctx.invoked_subcommand is None:
         from klaude_code.log import log
 
-        if vanilla and banana:
-            log(("Error: --banana cannot be combined with --vanilla", "red"))
-            raise typer.Exit(2)
-
         resume_by_id_value = resume_by_id.strip() if resume_by_id is not None else None
         if resume_by_id_value == "":
             log(("Error: --resume <id> cannot be empty", "red"))
@@ -249,18 +238,7 @@ def main_callback(
         update_terminal_title()
 
         chosen_model = model
-        if banana:
-            keywords = ["gemini-3-pro-image", "gemini-2.5-flash-image"]
-            model_result = select_model_interactive(keywords=keywords)
-            if model_result.status == ModelSelectStatus.SELECTED and model_result.model is not None:
-                chosen_model = model_result.model
-            elif model_result.status == ModelSelectStatus.CANCELLED:
-                return
-            else:
-                log(("Error: no available nano-banana model", "red"))
-                log(("Hint: set OPENROUTER_API_KEY or GOOGLE_API_KEY to enable nano-banana models", "yellow"))
-                raise typer.Exit(2)
-        elif model or select_model:
+        if model or select_model:
             model_result = select_model_interactive(preferred=model)
             if model_result.status == ModelSelectStatus.SELECTED and model_result.model is not None:
                 chosen_model = model_result.model

{klaude_code-2.19.0 → klaude_code-2.21.0}/src/klaude_code/config/assets/builtin_config.yaml

@@ -42,28 +42,19 @@ provider_list:
       reasoning_summary: concise
     cost: { input: 1.75, output: 14, cache_read: 0.17 }
 
-  - model_name: gpt-5.2-medium
-    model_id: gpt-5.2
-    context_limit: 400000
-    verbosity: high
+  - model_name: gpt-5.3-codex
+    model_id: gpt-5.3-codex
     thinking:
-      reasoning_effort: medium
-      reasoning_summary: concise
-    cost: { input: 1.75, output: 14, cache_read: 0.17 }
-
-  - model_name: gpt-5.2-low
-    model_id: gpt-5.2
+      reasoning_effort: high
+      reasoning_summary: auto
     context_limit: 400000
-    verbosity: low
-    thinking:
-      reasoning_effort: low
-      reasoning_summary: concise
+    max_tokens: 128000
     cost: { input: 1.75, output: 14, cache_read: 0.17 }
 
-  - model_name: gpt-5.2-codex
-    model_id: gpt-5.2-codex
+  - model_name: gpt-5.3-codex-xhigh
+    model_id: gpt-5.3-codex
     thinking:
-      reasoning_effort: high
+      reasoning_effort: xhigh
       reasoning_summary: auto
     context_limit: 400000
     max_tokens: 128000
@@ -74,8 +65,8 @@ provider_list:
   api_key: ${OPENROUTER_API_KEY}
   model_list:
 
-  - model_name: gpt-5.2-codex
-    model_id: gpt-5.2-codex
+  - model_name: gpt-5.3-codex
+    model_id: openai/gpt-5.3-codex
     thinking:
       reasoning_effort: high
       reasoning_summary: auto
@@ -83,23 +74,22 @@ provider_list:
     max_tokens: 128000
     cost: { input: 1.75, output: 14, cache_read: 0.17 }
 
-  - model_name: gpt-5.2-high
-    model_id: openai/gpt-5.2
-    max_tokens: 128000
-    context_limit: 400000
-    verbosity: high
+  - model_name: gpt-5.3-codex-xhigh
+    model_id: openai/gpt-5.3-codex
     thinking:
-      reasoning_effort: high
-      reasoning_summary: concise
+      reasoning_effort: xhigh
+      reasoning_summary: auto
+    context_limit: 400000
+    max_tokens: 128000
     cost: { input: 1.75, output: 14, cache_read: 0.17 }
 
-  - model_name: gpt-5.2-medium
+  - model_name: gpt-5.2-high
     model_id: openai/gpt-5.2
     max_tokens: 128000
     context_limit: 400000
     verbosity: high
     thinking:
-      reasoning_effort: medium
+      reasoning_effort: high
       reasoning_summary: concise
     cost: { input: 1.75, output: 14, cache_read: 0.17 }
 
@@ -176,32 +166,6 @@ provider_list:
       reasoning_effort: medium
     cost: { input: 0.5, output: 3, cache_read: 0.05 }
 
-  - model_name: nano-banana-pro
-    model_id: google/gemini-3-pro-image-preview
-    context_limit: 66000
-    modalities:
-    - image
-    - text
-    image_config:
-      image_size: "4K"
-    cost: { input: 2, output: 12, cache_read: 0.2, image: 120 }
-
-  - model_name: nano-banana
-    model_id: google/gemini-2.5-flash-image
-    context_limit: 33000
-    modalities:
-    - image
-    - text
-    cost: { input: 0.3, output: 2.5, cache_read: 0.03, image: 30 }
-
-  - model_name: aurora
-    model_id: openrouter/aurora-alpha
-    context_limit: 128000
-    thinking:
-      type: enabled
-      budget_tokens: 2048
-    cost: { input: 0, output: 0 }
-
   - model_name: grok
     model_id: x-ai/grok-4.1-fast
     context_limit: 2000000
@@ -230,22 +194,6 @@ provider_list:
       - z-ai
     cost: { input: 1, output: 3.2, cache_read: 0.1 }
 
-  - model_name: seedream
-    model_id: bytedance-seed/seedream-4.5
-    context_limit: 4000
-    cost: { input: 0, output: 9.581, image: 9.581 }
-    modalities:
-    - image
-    - text
-
-  - model_name: flux
-    model_id: black-forest-labs/flux.2-max
-    context_limit: 47000
-    cost: { input: 7.32, output: 7.32, image: 7.32 }
-    modalities:
-    - image
-    - text
-
 - provider_name: google
   protocol: google
   api_key: ${GOOGLE_API_KEY}
@@ -265,24 +213,6 @@ provider_list:
       reasoning_effort: medium
     cost: { input: 0.5, output: 3, cache_read: 0.05 }
 
-  - model_name: nano-banana-pro
-    model_id: gemini-3-pro-image-preview
-    context_limit: 66000
-    modalities:
-    - image
-    - text
-    image_config:
-      image_size: "4K"
-    cost: { input: 2, output: 12, cache_read: 0.2, image: 120 }
-
-  - model_name: nano-banana
-    model_id: gemini-2.5-flash-image
-    context_limit: 33000
-    modalities:
-    - image
-    - text
-    cost: { input: 0.3, output: 2.5, cache_read: 0.03, image: 30 }
-
 - provider_name: google-vertex
   protocol: google_vertex
   google_application_credentials: ${GOOGLE_APPLICATION_CREDENTIALS}
@@ -304,24 +234,6 @@ provider_list:
       reasoning_effort: medium
     cost: { input: 0.5, output: 3, cache_read: 0.05 }
 
-  - model_name: nano-banana-pro
-    model_id: gemini-3-pro-image-preview
-    context_limit: 66000
-    modalities:
-    - image
-    - text
-    image_config:
-      image_size: "4K"
-    cost: { input: 2, output: 12, cache_read: 0.2, image: 120 }
-
-  - model_name: nano-banana
-    model_id: gemini-2.5-flash-image
-    context_limit: 33000
-    modalities:
-    - image
-    - text
-    cost: { input: 0.3, output: 2.5, cache_read: 0.03, image: 30 }
-
 - provider_name: bedrock
   protocol: bedrock
   aws_access_key: ${AWS_ACCESS_KEY_ID}
@@ -441,6 +353,15 @@ provider_list:
     max_tokens: 128000
     cost: { input: 1.75, output: 14, cache_read: 0.17 }
 
+  - model_name: gpt-5.3-codex-xhigh
+    model_id: gpt-5.3-codex
+    thinking:
+      reasoning_effort: xhigh
+      reasoning_summary: auto
+    context_limit: 400000
+    max_tokens: 128000
+    cost: { input: 1.75, output: 14, cache_read: 0.17 }
+
 - provider_name: copilot
   protocol: copilot_oauth
   model_list:
@@ -454,6 +375,15 @@ provider_list:
     max_tokens: 128000
     cost: { input: 1.75, output: 14, cache_read: 0.17 }
 
+  - model_name: gpt-5.3-codex-xhigh
+    model_id: gpt-5.3-codex
+    thinking:
+      reasoning_effort: xhigh
+      reasoning_summary: auto
+    context_limit: 400000
+    max_tokens: 128000
+    cost: { input: 1.75, output: 14, cache_read: 0.17 }
+
   - model_name: sonnet
     model_id: claude-sonnet-4-6
     context_limit: 200000
@@ -517,4 +447,3 @@ compact_model: gemini-flash
 
 sub_agent_models:
   Explore: sonnet-no-thinking
-  Web: sonnet-no-thinking

{klaude_code-2.19.0 → klaude_code-2.21.0}/src/klaude_code/config/config.py

@@ -383,20 +383,6 @@ class Config(BaseModel):
             if include_disabled or not model.disabled
         ]
 
-    def has_available_image_model(self) -> bool:
-        """Check if any image generation model is available."""
-        for entry in self.iter_model_entries(only_available=True, include_disabled=False):
-            if entry.modalities and "image" in entry.modalities:
-                return True
-        return False
-
-    def get_first_available_image_model(self) -> str | None:
-        """Get the first available image generation model, or None."""
-        for entry in self.iter_model_entries(only_available=True, include_disabled=False):
-            if entry.modalities and "image" in entry.modalities:
-                return entry.model_name
-        return None
-
     async def save(self) -> None:
         """Save user config to file (excludes builtin providers).
 
@@ -444,7 +430,7 @@ def get_example_config() -> UserConfig:
     return UserConfig(
         main_model="opus",
         compact_model="gemini-flash",
-        sub_agent_models={"Explore": "haiku", "Web": "sonnet", "Task": "sonnet"},
+        sub_agent_models={"Explore": "haiku", "Task": "sonnet"},
         provider_list=[
             UserProviderConfig(
                 provider_name="my-provider",

klaude_code-2.21.0/src/klaude_code/config/sub_agent_model_helper.py

@@ -0,0 +1,110 @@
+"""Helper for sub-agent model availability and selection logic."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import TYPE_CHECKING
+
+from klaude_code.protocol import tools
+from klaude_code.protocol.sub_agent import (
+    SubAgentProfile,
+    get_sub_agent_profile,
+    iter_sub_agent_profiles,
+)
+from klaude_code.protocol.tools import SubAgentType
+
+if TYPE_CHECKING:
+    from klaude_code.config.config import Config, ModelEntry
+
+
+@dataclass
+class SubAgentModelInfo:
+    """Sub-agent and its current model configuration."""
+
+    profile: SubAgentProfile
+    # Explicitly configured model selector (from config), if any.
+    configured_model: str | None
+
+    # Effective model name used by this sub-agent.
+    # - When configured_model is set: equals configured_model.
+    # - When inheriting from defaults: resolved model name.
+    effective_model: str | None
+
+
+@dataclass(frozen=True, slots=True)
+class EmptySubAgentModelBehavior:
+    """Human-facing description for an unset (empty) sub-agent model config."""
+
+    # Summary text for UI (kept UI-framework agnostic).
+    description: str
+
+    # Best-effort resolved model name (if any).
+    resolved_model_name: str | None
+
+
+class SubAgentModelHelper:
+    """Centralized logic for sub-agent availability and model selection."""
+
+    def __init__(self, config: Config) -> None:
+        self._config = config
+
+    def describe_empty_model_config_behavior(
+        self,
+        sub_agent_type: str,
+        *,
+        main_model_name: str,
+    ) -> EmptySubAgentModelBehavior:
+        """Describe what happens when a sub-agent model is not configured.
+
+        Sub-agents default to the Task model if configured, otherwise
+        they inherit the main model.
+        """
+
+        _ = get_sub_agent_profile(sub_agent_type)
+        task_model = self._config.sub_agent_models.get(tools.TASK)
+        resolved = task_model or main_model_name
+        return EmptySubAgentModelBehavior(
+            description=f"use default behavior: {resolved}",
+            resolved_model_name=resolved,
+        )
+
+    def get_available_sub_agents(self) -> list[SubAgentModelInfo]:
+        """Return all available sub-agents with their current model config.
+
+        For sub-agents without explicit config, resolve from Task/main defaults.
+        """
+        result: list[SubAgentModelInfo] = []
+        for profile in iter_sub_agent_profiles():
+            configured_model = self._config.sub_agent_models.get(profile.name)
+            effective_model = (
+                configured_model or self._config.sub_agent_models.get(tools.TASK) or self._config.main_model
+            )
+            result.append(
+                SubAgentModelInfo(
+                    profile=profile,
+                    configured_model=configured_model,
+                    effective_model=effective_model,
+                )
+            )
+        return result
+
+    def get_selectable_models(self, sub_agent_type: str) -> list[ModelEntry]:
+        """Return selectable models for a specific sub-agent type.
+
+        Returns all available models.
+        """
+        _ = get_sub_agent_profile(sub_agent_type)
+        all_models = self._config.iter_model_entries(only_available=True, include_disabled=False)
+        return all_models
+
+    def build_sub_agent_client_configs(self) -> dict[SubAgentType, str]:
+        """Return model names for each sub-agent that needs a dedicated client."""
+        result: dict[SubAgentType, str] = {}
+        for profile in iter_sub_agent_profiles():
+            model_name = self._config.sub_agent_models.get(profile.name)
+            if model_name:
+                result[profile.name] = model_name
+        task_model = self._config.sub_agent_models.get(tools.TASK)
+        if task_model:
+            result.setdefault(tools.TASK, task_model)
+        return result

{klaude_code-2.19.0 → klaude_code-2.21.0}/src/klaude_code/config/thinking.py

@@ -10,7 +10,6 @@ from typing import Literal
 from klaude_code.protocol import llm_param
 from klaude_code.protocol.model_id import (
     is_claude_model_any,
-    is_codex_max_model,
     is_gemini_flash_model,
     is_gpt51_model,
     is_gpt52_model,
@@ -24,7 +23,6 @@ ReasoningEffort = Literal["high", "medium", "low", "minimal", "none", "xhigh"]
 RESPONSES_LEVELS = ["low", "medium", "high"]
 RESPONSES_GPT51_LEVELS = ["none", "low", "medium", "high"]
 RESPONSES_GPT52_LEVELS = ["none", "low", "medium", "high", "xhigh"]
-RESPONSES_CODEX_MAX_LEVELS = ["medium", "high", "xhigh"]
 RESPONSES_GEMINI_FLASH_LEVELS = ["minimal", "low", "medium", "high"]
 
 ANTHROPIC_LEVELS: list[tuple[str, int | None]] = [
@@ -37,8 +35,6 @@ ANTHROPIC_LEVELS: list[tuple[str, int | None]] = [
 
 def get_levels_for_responses(model_name: str | None) -> list[str]:
     """Get thinking levels for responses protocol."""
-    if is_codex_max_model(model_name):
-        return RESPONSES_CODEX_MAX_LEVELS
     if is_gpt52_model(model_name):
         return RESPONSES_GPT52_LEVELS
     if is_gpt51_model(model_name):

{klaude_code-2.19.0 → klaude_code-2.21.0}/src/klaude_code/const.py

@@ -51,8 +51,6 @@ OPENROUTER_BASE_URL = "https://openrouter.ai/api/v1"  # OpenRouter API base URL
 CODEX_BASE_URL = "https://chatgpt.com/backend-api/codex"  # Codex API base URL
 CODEX_USER_AGENT = "pi"  # Codex user agent string
 
-SUPPORTED_IMAGE_SIZES = {"1K", "2K", "4K"}  # Supported image sizes for ImageGen tool
-
 LOW_CACHE_HIT_RATE_THRESHOLD = 0.9  # Cache hit rate below this triggers a warning
 THROUGHPUT_MIN_DURATION_SEC = 0.15  # Minimum duration (seconds) for throughput calculation
 INITIAL_RETRY_DELAY_S = 1.0  # Initial delay before retrying a failed turn (seconds)
@@ -82,7 +80,6 @@ READ_CHAR_LIMIT_PER_LINE = 2000  # Maximum characters per line before truncation
 READ_GLOBAL_LINE_CAP = _get_int_env("KLAUDE_READ_GLOBAL_LINE_CAP", 2000)  # Maximum lines to read from a file
 READ_MAX_CHARS = _get_int_env("KLAUDE_READ_MAX_CHARS", 50000)  # Maximum total characters to read
 READ_MAX_IMAGE_BYTES = _get_int_env("KLAUDE_READ_MAX_IMAGE_BYTES", 64 * 1024 * 1024)  # Max image size (64MB)
-IMAGE_OUTPUT_MAX_BYTES = _get_int_env("KLAUDE_IMAGE_OUTPUT_MAX_BYTES", 64 * 1024 * 1024)  # Max decoded image (64MB)
 BINARY_CHECK_SIZE = 8192  # Bytes to check for binary file detection
 
 
@@ -102,13 +99,12 @@ BASH_MODE_SESSION_OUTPUT_MAX_BYTES = 200 * 1024 * 1024  # Max command output cap
 WEB_FETCH_DEFAULT_TIMEOUT_SEC = 30  # Default timeout for web fetch requests (seconds)
 WEB_FETCH_USER_AGENT = "Mozilla/5.0 (compatible; KlaudeCode/1.0)"  # User-Agent header for web requests
 WEB_FETCH_MAX_RESPONSE_BYTES = 5 * 1024 * 1024  # Maximum response body size (5MB)
-WEB_FETCH_MAX_REDIRECTS = 3  # Maximum number of redirects to follow
+WEB_FETCH_MAX_REDIRECTS = 10  # Maximum number of redirects to follow
 URL_FILENAME_MAX_LENGTH = 80  # Maximum length for extracting filename from URL
 WEB_SEARCH_DEFAULT_MAX_RESULTS = 10  # Default number of search results
 WEB_SEARCH_MAX_RESULTS_LIMIT = 20  # Maximum number of search results allowed
 WEB_CACHE_TTL_SECONDS = 900  # TTL for web search/fetch cache (15 minutes)
 WEB_CACHE_MAX_ENTRIES = 100  # Maximum entries in web cache
-MERMAID_LIVE_PREFIX = "https://mermaid.live/view#pako:"  # Mermaid.live URL prefix
 
 
 # =============================================================================
@@ -250,3 +246,7 @@ class ProjectPaths:
 
     def meta_file(self, session_id: str) -> Path:
         return self.session_dir(session_id) / "meta.json"
+
+    @property
+    def memory_dir(self) -> Path:
+        return self.base_dir / "memory"