mycode-cli 0.1.2__tar.gz → 0.2.1__tar.gz

mycode_cli-0.2.1/.basedpyright/baseline.json

@@ -0,0 +1,3 @@
+{
+    "files": {}
+}

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/.github/workflows/ci.yml

@@ -21,14 +21,14 @@ jobs:
         uses: astral-sh/setup-uv@v7
 
       - name: Install pnpm
-        uses: pnpm/action-setup@v4
+        uses: pnpm/action-setup@v5
         with:
           version: 10
 
       - name: Install Node.js
         uses: actions/setup-node@v6
         with:
-          node-version: 20
+          node-version: 24
           cache: pnpm
           cache-dependency-path: web/pnpm-lock.yaml
 
@@ -41,6 +41,9 @@ jobs:
       - name: Lint Python
         run: uv run ruff check .
 
+      - name: Typecheck Python
+        run: uv run basedpyright
+
       - name: Run Python tests
         run: uv run pytest
 
@@ -51,14 +54,14 @@ jobs:
         uses: actions/checkout@v6
 
       - name: Install pnpm
-        uses: pnpm/action-setup@v4
+        uses: pnpm/action-setup@v5
         with:
           version: 10
 
       - name: Install Node.js
         uses: actions/setup-node@v6
         with:
-          node-version: 20
+          node-version: 24
           cache: pnpm
           cache-dependency-path: web/pnpm-lock.yaml
 
@@ -84,14 +87,14 @@ jobs:
         uses: astral-sh/setup-uv@v7
 
       - name: Install pnpm
-        uses: pnpm/action-setup@v4
+        uses: pnpm/action-setup@v5
         with:
           version: 10
 
       - name: Install Node.js
         uses: actions/setup-node@v6
         with:
-          node-version: 20
+          node-version: 24
           cache: pnpm
           cache-dependency-path: web/pnpm-lock.yaml
 

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/.github/workflows/publish.yml

@@ -22,14 +22,14 @@ jobs:
         uses: astral-sh/setup-uv@v7
 
       - name: Install pnpm
-        uses: pnpm/action-setup@v4
+        uses: pnpm/action-setup@v5
         with:
           version: 10
 
       - name: Install Node.js
         uses: actions/setup-node@v6
         with:
-          node-version: 20
+          node-version: 24
           cache: pnpm
           cache-dependency-path: web/pnpm-lock.yaml
 

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/.pre-commit-config.yaml

@@ -12,6 +12,13 @@ repos:
       - id: check-merge-conflict
       - id: detect-private-key
 
+  # Markdown
+  - repo: https://github.com/igorshubovych/markdownlint-cli
+    rev: v0.48.0
+    hooks:
+      - id: markdownlint
+        args: ["--dot", "--fix", "--disable", "MD013", "--"]
+
   # Python - Ruff (linter + formatter)
   - repo: https://github.com/astral-sh/ruff-pre-commit
     rev: v0.14.13

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/AGENTS.md

@@ -168,11 +168,14 @@ All adapters implement `ProviderAdapter.stream_turn()`. Message projection to pr
 
 ```bash
 uv sync --dev                                          # Python setup
+uv run basedpyright                                    # Python type checking
+uv run pytest                                          # Python tests
 uv run mycode                                          # run CLI
 uv run mycode web --dev                                # API only (backend for Vite dev)
+pnpm --dir web typecheck                               # Web UI type checking
 pnpm --dir web test:run                                # run web UI tests once
 pnpm --dir web dev                                     # Vite web UI dev server
-uv run --no-project python scripts/build_web.py       # rebuild packaged web UI
+uv run --no-project python scripts/build_web.py        # rebuild packaged web UI
 uv build                                               # build wheel + sdist
 ```
 

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mycode-cli
-Version: 0.1.2
+Version: 0.2.1
 Summary: Minimal coding agent with web UI
 Project-URL: Homepage, https://github.com/legibet/mycode
 Project-URL: Repository, https://github.com/legibet/mycode
@@ -41,7 +41,7 @@ A minimal coding agent. Inspired by [pi](https://github.com/badlogic/pi-mono).
 - Unified message format and robust cross-provider replay.
 - 4 built-in tools: `read`, `write`, `edit`, `bash`, expanded via skills.
 - Inspectable runtime, append-only JSONL sessions.
-- Native image input.
+- Native image and pdf input support.
 - Mobile-friendly web UI.
 
 ## Quick Start
@@ -142,7 +142,7 @@ Config is loaded from `~/.mycode/config.json` (global) and `<workspace>/.mycode/
 
 ## CLI Reference
 
-```
+```bash
 mycode                            start interactive session (new)
 mycode --continue                 resume the most recent session
 mycode --session <id>             resume a specific session

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/README.md

@@ -8,7 +8,7 @@ A minimal coding agent. Inspired by [pi](https://github.com/badlogic/pi-mono).
 - Unified message format and robust cross-provider replay.
 - 4 built-in tools: `read`, `write`, `edit`, `bash`, expanded via skills.
 - Inspectable runtime, append-only JSONL sessions.
-- Native image input.
+- Native image and pdf input support.
 - Mobile-friendly web UI.
 
 ## Quick Start
@@ -109,7 +109,7 @@ Config is loaded from `~/.mycode/config.json` (global) and `<workspace>/.mycode/
 
 ## CLI Reference
 
-```
+```bash
 mycode                            start interactive session (new)
 mycode --continue                 resume the most recent session
 mycode --session <id>             resume a specific session

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/docs/api.md

@@ -36,16 +36,23 @@ Structured `input` uses `ChatInputBlock`:
 ```json
 [
   {"type": "text", "text": "describe this"},
+  {"type": "text", "text": "print('hi')", "name": "main.py", "is_attachment": true},
   {"type": "image", "path": "cat.png"},
-  {"type": "image", "data": "<base64>", "mime_type": "image/png", "name": "cat.png"}
+  {"type": "image", "data": "<base64>", "mime_type": "image/png", "name": "cat.png"},
+  {"type": "document", "path": "report.pdf"},
+  {"type": "document", "data": "<base64>", "mime_type": "application/pdf", "name": "report.pdf"}
 ]
 ```
 
 - `type: "text"` — uses `text`
+- `type: "text"` with `is_attachment=true` — wraps UTF-8 file content as the same `<file ...>` attachment text used by CLI `@file`
 - `type: "image"` — uses `path` or inline base64 `data`
+- `type: "document"` — uses `path` or inline base64 `data`
 - `mime_type` is required when `data` is provided
 - `path` accepts `image/png`, `image/jpeg`, `image/gif`, `image/webp`
+- `path` accepts `application/pdf` for `type: "document"`
 - The resolved model must have `supports_image_input=true`
+- The resolved model must have `supports_pdf_input=true` for `type: "document"`
 
 Response:
 
@@ -95,7 +102,9 @@ Response:
       "reasoning_models": ["claude-sonnet-4-6"],
       "reasoning_effort": "auto",
       "supports_image_input": true,
-      "image_input_models": ["claude-sonnet-4-6"]
+      "image_input_models": ["claude-sonnet-4-6"],
+      "supports_pdf_input": true,
+      "pdf_input_models": ["claude-sonnet-4-6"]
     }
   },
   "default": { "provider": "<provider_name>", "model": "claude-sonnet-4-6" },
@@ -107,7 +116,7 @@ Response:
 }
 ```
 
-`reasoning_models` is returned only when `supports_reasoning_effort` is true. `image_input_models` lists models with `supports_image_input=true`.
+`reasoning_models` is returned only when `supports_reasoning_effort` is true. `image_input_models` lists models with `supports_image_input=true`. `pdf_input_models` lists models with `supports_pdf_input=true`.
 
 ## Sessions
 

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/docs/config.md

@@ -31,7 +31,8 @@ Config resolution: `get_settings(cwd)` → returns `Settings` dataclass.
           "context_window": 400000,
           "max_output_tokens": 128000,
           "supports_reasoning": true,
-          "supports_image_input": true
+          "supports_image_input": true,
+          "supports_pdf_input": true
         },
         "model-b": {}
       },
@@ -55,6 +56,7 @@ Config resolution: `get_settings(cwd)` → returns `Settings` dataclass.
 - `providers.<name>.models.<model>.max_output_tokens` — override the provider output limit
 - `providers.<name>.models.<model>.supports_reasoning` — override whether reasoning effort is available
 - `providers.<name>.models.<model>.supports_image_input` — override image input support
+- `providers.<name>.models.<model>.supports_pdf_input` — override PDF input support
 - `providers.<name>.api_key` — literal value or `${ENV_NAME}` reference
 - `providers.<name>.base_url` — override the adapter's default base URL
 - `providers.<name>.reasoning_effort` — per-provider override of the global default
@@ -104,6 +106,7 @@ Options: `auto` (default) · `none` · `low` · `medium` · `high` · `xhigh`
 
 - `supports_reasoning` — whether the model supports extended thinking
 - `supports_image_input` — whether the model accepts image input
+- `supports_pdf_input` — whether the model accepts PDF input
 - `context_window` — used for compact threshold calculation
 - `max_output_tokens` — passed to the provider as the output limit; defaults to `16384` when not available
 - `context_window` — defaults to `128000` when the bundled catalog does not provide one

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/docs/providers.md

@@ -19,17 +19,20 @@ class ProviderAdapter(ABC):
 ```
 
 `prepare_messages()` converts canonical session history to provider-safe wire format. The base implementation in `base.py` handles:
+
 - Stripping error/aborted/cancelled assistant turns
 - Projecting tool call IDs (some providers restrict charset/length)
-- Dropping replay images when `request.supports_image_input` is false
+- Replacing replay images with a short text notice when `request.supports_image_input` is false
+- Replacing replay PDFs with a short text notice when `request.supports_pdf_input` is false
 - Flushing interrupted tool calls with synthetic error results
 
 `stream_turn()` yields `ProviderStreamEvent` objects:
+
 - `thinking_delta` — reasoning text
 - `text_delta` — response text
 - `message_done` — final `ConversationMessage` with all blocks and metadata
 
-`ProviderRequest` carries: provider, model, session_id, messages, system, tools, max_tokens, api_key, api_base, reasoning_effort, supports_image_input.
+`ProviderRequest` carries: provider, model, session_id, messages, system, tools, max_tokens, api_key, api_base, reasoning_effort, supports_image_input, supports_pdf_input.
 
 ## Adapters
 
@@ -46,6 +49,7 @@ class ProviderAdapter(ABC):
 - Adds ephemeral `cache_control` to system prompt block and last user content block
 - Tool call IDs projected to ASCII-safe format (letters, numbers, underscores, dashes, max 64 chars) with SHA1 collision suffix
 - Images serialize as Anthropic `image` blocks with base64 `source`
+- PDFs serialize as Anthropic `document` blocks with base64 `source`
 
 ### `moonshotai` — `anthropic_like.py`
 
@@ -57,6 +61,7 @@ class ProviderAdapter(ABC):
 - Prior reasoning must be replayed on later tool-loop turns when thinking is enabled
 - Shares Anthropic-like ephemeral cache markers and tool call ID projection
 - Same image format as `anthropic`
+- Same PDF format as `anthropic`
 
 ### `minimax` — `anthropic_like.py`
 
@@ -68,6 +73,7 @@ class ProviderAdapter(ABC):
 - Preserves provider-native thinking signatures in `block.meta.native`
 - Shares Anthropic-like ephemeral cache markers and tool call ID projection
 - Same image format as `anthropic`
+- Same PDF format as `anthropic`
 
 ### `google` — `gemini.py`
 
@@ -87,6 +93,7 @@ class ProviderAdapter(ABC):
 - Gemini validates function_call id/name match between function_call and function_response pairs
 - `thinking_config.include_thoughts` always true; effort level controls `thinking_level`
 - Images serialize as `inline_data`
+- PDFs serialize as `inline_data`
 
 ### `openai` — `openai_responses.py`
 
@@ -97,11 +104,12 @@ class ProviderAdapter(ABC):
 - Default models: `gpt-5.4`, `gpt-5.4-mini`
 - `supports_reasoning_effort`: true (`reasoning = {"effort": ...}`, values: `none/low/medium/high/xhigh`)
 - Runs stateless: `store=false`, `include=["reasoning.encrypted_content"]`
-- Native `response.output` items persisted under `assistant.meta.native.output_items` and replayed directly
+- Streaming turns persist completed output items from `response.output_item.done` under `assistant.meta.native.output_items` and replay them directly
 - Tool results replay as `function_call_output`; foreign thinking never converted to OpenAI reasoning items
 - Passes `prompt_cache_key` using current session id
 - Tool schemas use `strict: true` with nullable optional parameters
 - Images serialize as `input_image`
+- PDFs serialize as `input_file`
 
 ### `openai_chat` — `openai_chat.py`
 
@@ -113,6 +121,7 @@ class ProviderAdapter(ABC):
 - Preserves third-party reasoning extensions (`reasoning_content`, `reasoning_details`) from SDK extras
 - Sends `stream_options: {include_usage: true}`
 - Images serialize as `image_url` parts with data URLs
+- PDFs serialize as `file` parts with base64 data URLs
 
 ### `deepseek` — `openai_chat.py`
 
@@ -143,6 +152,7 @@ class ProviderAdapter(ABC):
 - `supports_reasoning_effort`: true (forwarded through `extra_body.reasoning.effort`)
 - `auto_discoverable`: true
 - Same image format as `openai_chat`
+- Same PDF format as `openai_chat`
 
 ## Reasoning Effort Mapping
 
@@ -163,7 +173,8 @@ Config-resolved `reasoning_effort` is only applied when both `adapter.supports_r
 1. Skip assistant messages with `stop_reason` in `{error, aborted, cancelled}`
 2. Project tool call IDs to provider-safe format (only Anthropic-like adapters override this)
 3. Preserve `block.meta.native` for provider-specific replay data (signatures, output items, part metadata)
-4. Drop replay images when `request.supports_image_input` is false
-5. Insert synthetic error tool results when pending tool calls would otherwise make replay invalid
+4. Replace replay images with a short text notice when `request.supports_image_input` is false
+5. Replace replay PDFs with a short text notice when `request.supports_pdf_input` is false
+6. Insert synthetic error tool results when pending tool calls would otherwise make replay invalid
 
 Provider-specific replay logic lives inside each adapter's serialization methods (e.g., Gemini's `_build_contents` replays native `Part` metadata, OpenAI's `_native_output_items` replays stored output items). These run after `prepare_messages()` produces the canonical replay transcript.

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/docs/sessions.md

@@ -4,7 +4,7 @@ Source: `mycode/core/session.py`
 
 ## Storage Layout
 
-```
+```text
 ~/.mycode/sessions/<session_id>/
   meta.json        # session metadata
   messages.jsonl   # one JSON record per line (append-only)

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/docs/web.md

@@ -11,7 +11,7 @@ CORS is enabled for all origins in the FastAPI app.
 
 ## Component Structure
 
-```
+```text
 web/src/
   App.tsx                # root layout, config loading, session init
   main.tsx               # React entry
@@ -52,11 +52,13 @@ web/src/
 ## Message State Model
 
 `useChat.ts` stores three related pieces of state:
+
 - `rawMessages` — canonical block messages
 - `messages` — render-ready messages
 - `toolRuntimeById` — ephemeral tool runtime state
 
 State is managed via `useReducer` with actions:
+
 - `set_messages` — load session history from server
 - `start_turn` — optimistic user message + empty assistant
 - `rewind_and_start_turn` — rewind + optimistic new turn
@@ -65,11 +67,13 @@ State is managed via `useReducer` with actions:
 `buildRenderMessages()` in `utils/messages.ts` is used when loading or rebuilding from canonical messages. During streaming, the reducer updates both `rawMessages` and `messages` incrementally.
 
 Key design decisions:
+
 - Tool results persisted as `user` messages with `tool_result` blocks are visually folded into the preceding assistant message during rendering
 - Each render message and block gets a stable `renderKey` for React reconciliation
 - `sourceIndex` tracks the original message position for scroll targeting
 
 Rendering rules:
+
 - `thinking` blocks → `ReasoningBlock` (expanded while streaming, auto-collapses after)
 - `tool_use` blocks → `ToolCard` (with matching `tool_result` and live runtime folded in)
 - `text` blocks → `MarkdownBlock`
@@ -85,14 +89,17 @@ Rendering rules:
 6. 409 conflict: attach to the existing run's stream
 
 Streaming state tracking:
+
 - `streamTokenRef` — incremented to invalidate stale streams
 - `pendingRequestTokenRef` — deduplicates concurrent send requests
 - `activeRunRef` — tracks the current run for cancel
 
-Image input:
-- `InputArea` supports file picker and drag-and-drop
-- Images are sent as structured `input` blocks
-- The attachment button uses `image_input_models`; pending images are cleared on unsupported model switch
+Attachments:
+
+- `InputArea` always shows the attachment button and supports file picker and drag-and-drop
+- UTF-8 text/code/config files are attached as the same text snapshot format used by CLI `@file`
+- Images and PDFs are sent as structured `input` blocks
+- The attachment button uses `image_input_models` and `pdf_input_models`; unsupported pending attachments are cleared on model switch
 
 ## Config Persistence
 

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/hatch_build.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import sys
 from pathlib import Path
+from typing import Any
 
 from hatchling.builders.hooks.plugin.interface import BuildHookInterface
 
@@ -16,8 +17,8 @@ def build_web_assets() -> None:
     build_web()
 
 
-class CustomBuildHook(BuildHookInterface):
-    def initialize(self, version: str, build_data: dict) -> None:
+class CustomBuildHook(BuildHookInterface[Any]):
+    def initialize(self, version: str, build_data: dict[str, Any]) -> None:
         del version, build_data
         build_web_assets()
 

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/mycode/cli/chat.py

@@ -7,15 +7,18 @@ import html
 import re
 import shlex
 from base64 import b64encode
+from collections.abc import Iterable
 from pathlib import Path
-from typing import Any
+from typing import Any, override
 
 from prompt_toolkit import PromptSession
 from prompt_toolkit.application import Application, get_app
-from prompt_toolkit.completion import Completer, Completion
-from prompt_toolkit.formatted_text import ANSI
+from prompt_toolkit.completion import CompleteEvent, Completer, Completion
+from prompt_toolkit.document import Document
+from prompt_toolkit.formatted_text import ANSI, StyleAndTextTuples
 from prompt_toolkit.history import FileHistory
 from prompt_toolkit.key_binding import KeyBindings
+from prompt_toolkit.key_binding.key_processor import KeyPressEvent
 from prompt_toolkit.keys import Keys
 from prompt_toolkit.layout import Layout
 from prompt_toolkit.widgets import RadioList
@@ -23,9 +26,9 @@ from rich.text import Text
 
 from mycode.core.agent import Agent
 from mycode.core.config import resolve_mycode_home
-from mycode.core.messages import build_message, image_block, text_block
+from mycode.core.messages import build_message, document_block, image_block, text_block
 from mycode.core.session import SessionStore
-from mycode.core.tools import detect_image_mime_type, resolve_path
+from mycode.core.tools import detect_document_mime_type, detect_image_mime_type, resolve_path
 
 from .render import ReplyRenderer, TerminalView, format_local_timestamp
 from .runtime import (
@@ -62,17 +65,19 @@ _AT_PATH_RE = re.compile(r"(?<!\S)@(?:(?P<quote>['\"])(?P<quoted>[^'\"]*)|(?P<pl
 _FOCUSED_STYLE = "bold blue" if TERMINAL_THEME == "light" else "bold cyan"
 
 
-class _InlineRadioList[T](RadioList):
+class _InlineRadioList[T](RadioList[T]):
     """Arrow-key list that shows > on the focused item and exits on Enter."""
 
+    @override
     def _handle_enter(self) -> None:
         # Only called by Enter/Space (not arrows), so safe to exit.
         self.current_value = self.values[self._selected_index][0]
         get_app().exit(result=self.current_value)
 
-    def _get_text_fragments(self):
+    @override
+    def _get_text_fragments(self) -> StyleAndTextTuples:
         # Override rendering: show > based on focus, not checked state.
-        result: list[tuple[str, str]] = []
+        result: StyleAndTextTuples = []
         for i, (_value, text) in enumerate(self.values):
             focused = i == self._selected_index
             style = _FOCUSED_STYLE if focused else ""
@@ -97,7 +102,7 @@ async def choose[T](options: list[tuple[T, str]], *, default: T | None = None) -
 
     @kb.add("c-c")
     @kb.add("escape")
-    def _cancel(event) -> None:
+    def _cancel(event: KeyPressEvent) -> None:
         event.app.exit(result=None)
 
     app: Application[T | None] = Application(
@@ -116,7 +121,8 @@ class _PromptCompleter(Completer):
     def __init__(self, *, cwd: str | None = None) -> None:
         self._cwd = cwd
 
-    def get_completions(self, document, complete_event):
+    def get_completions(self, document: Document, complete_event: CompleteEvent) -> Iterable[Completion]:
+        del complete_event
         text_before_cursor = document.text_before_cursor
         text = text_before_cursor.lstrip()
         if self._cwd:
@@ -186,16 +192,25 @@ def _build_chat_key_bindings() -> KeyBindings:
     """Build key bindings for the main chat prompt."""
     kb = KeyBindings()
 
-    kb.add("c-l")(lambda event: event.app.renderer.clear())
+    def _clear(event: KeyPressEvent) -> None:
+        event.app.renderer.clear()
+
+    kb.add("c-l")(_clear)
 
     # In multiline mode the default Enter inserts a newline; override it to submit.
-    kb.add("enter", eager=True)(lambda event: event.current_buffer.validate_and_handle())
+    def _submit(event: KeyPressEvent) -> None:
+        event.current_buffer.validate_and_handle()
+
+    kb.add("enter", eager=True)(_submit)
 
     # Esc+Enter (Meta+Enter) inserts a newline for multiline input.
-    kb.add("escape", "enter")(lambda event: event.current_buffer.insert_text("\n"))
+    def _insert_newline(event: KeyPressEvent) -> None:
+        event.current_buffer.insert_text("\n")
+
+    kb.add("escape", "enter")(_insert_newline)
 
     @kb.add(Keys.BracketedPaste, eager=True)
-    def _handle_bracketed_paste(event) -> None:
+    def _handle_bracketed_paste(event: KeyPressEvent) -> None:
         pasted = event.data.replace("\r\n", "\n").replace("\r", "\n")
         event.current_buffer.insert_text(_rewrite_pasted_file_paths(pasted) or pasted)
 
@@ -225,7 +240,7 @@ class TerminalChat:
         self.store = store
         self.session_id = session_id
         self.view = view or TerminalView()
-        self.prompt_session = PromptSession(
+        self.prompt_session: PromptSession[str] = PromptSession(
             history=FileHistory(history_file_path()),
             completer=_PromptCompleter(cwd=self.agent.cwd),
             key_bindings=_build_chat_key_bindings(),
@@ -284,9 +299,9 @@ class TerminalChat:
     def _build_user_message(self, text: str) -> dict[str, Any]:
         """Build one user message with the raw prompt first, then resolved attachments.
 
-        Text files are appended as extra text blocks in their final provider-facing
-        form. Images are appended as image blocks. Only explicit `@path` tokens
-        that resolve to real files are attached.
+        Text files are appended as extra text blocks. Images and PDFs become
+        native blocks only when the current model supports that input type.
+        Only explicit `@path` tokens that resolve to real files are attached.
         """
 
         blocks = [text_block(text)]
@@ -309,10 +324,27 @@ class TerminalChat:
                 continue
             seen.add(path_text)
 
-            image_mime_type = detect_image_mime_type(path)
-            if image_mime_type:
-                image_data = b64encode(path.read_bytes()).decode("utf-8")
-                blocks.append(image_block(image_data, mime_type=image_mime_type, name=path.name))
+            # Detect image or document; bundle (kind, mime, supported) together.
+            media: tuple[str, str, bool] | None = None
+            if m := detect_image_mime_type(path):
+                media = ("image", m, self.agent.supports_image_input)
+            elif m := detect_document_mime_type(path):
+                media = ("document", m, self.agent.supports_pdf_input)
+
+            if media:
+                kind, mime_type, supported = media
+                if supported:
+                    data = b64encode(path.read_bytes()).decode("utf-8")
+                    fn = image_block if kind == "image" else document_block
+                    blocks.append(fn(data, mime_type=mime_type, name=path.name))
+                else:
+                    label = "image input" if kind == "image" else "PDF input"
+                    blocks.append(
+                        text_block(
+                            f'<file name="{html.escape(path_text, quote=True)}" media_type="{mime_type}" kind="{kind}">Current model does not support {label}.</file>',
+                            meta={"attachment": True, "path": path_text},
+                        )
+                    )
                 continue
 
             # Reuse the existing read tool so attached text files follow the same
@@ -459,20 +491,16 @@ class TerminalChat:
 
         # Collect real user text messages (skip synthetic compact summaries
         # and tool-result-only user messages).
-        user_turns: list[tuple[int, str]] = []  # (message_index, full_text)
+        user_turns: dict[int, str] = {}  # message_index -> text
         for i, msg in enumerate(messages):
             if msg.get("role") != "user":
                 continue
             if (msg.get("meta") or {}).get("synthetic"):
                 continue
-            blocks = msg.get("content") or []
-            text = ""
-            for b in blocks:
+            for b in msg.get("content") or []:
                 if isinstance(b, dict) and b.get("type") == "text" and b.get("text"):
-                    text = str(b["text"]).strip()
+                    user_turns[i] = str(b["text"]).strip()
                     break
-            if text:
-                user_turns.append((i, text))
 
         if not user_turns:
             self.view.console.print("[dim]no user messages to rewind to[/dim]")
@@ -480,7 +508,7 @@ class TerminalChat:
 
         # Build selector options — most recent first.
         options: list[tuple[int, str]] = []
-        for msg_index, text in reversed(user_turns):
+        for msg_index, text in reversed(list(user_turns.items())):
             preview = text.replace("\n", " ")[:60]
             if len(text) > 60:
                 preview += "..."
@@ -490,12 +518,7 @@ class TerminalChat:
         if selected is None:
             return None
 
-        # Look up the full text of the selected message for prefill.
-        original_text = ""
-        for msg_index, text in user_turns:
-            if msg_index == selected:
-                original_text = text
-                break
+        original_text = user_turns.get(selected, "")
 
         # Persist the rewind event and truncate in-memory messages.
         await self.store.append_rewind(self.session_id, selected)

{mycode_cli-0.1.2 → mycode_cli-0.2.1}/mycode/cli/main.py

@@ -11,6 +11,7 @@ import typer
 
 from mycode.core.agent import Agent
 from mycode.core.config import get_settings, resolve_provider
+from mycode.core.messages import ConversationMessage
 from mycode.core.session import SessionStore
 
 from .chat import TerminalChat
@@ -22,9 +23,6 @@ session_app = typer.Typer(help="Session management")
 app.add_typer(session_app, name="session")
 
 
-# -- Shared helpers ----------------------------------------------------------
-
-
 async def run_noninteractive(
     agent: Agent,
     *,
@@ -32,11 +30,11 @@ async def run_noninteractive(
     session_id: str,
     message: str,
 ) -> int:
-    """Run one CLI message and print only the final assistant reply."""
+    """Run one message non-interactively and print only the final assistant reply."""
 
-    latest_assistant: dict | None = None
+    latest_assistant: ConversationMessage | None = None
 
-    async def persist(payload: dict) -> None:
+    async def persist(payload: ConversationMessage) -> None:
         nonlocal latest_assistant
         if payload.get("role") == "assistant":
             latest_assistant = payload
@@ -67,7 +65,7 @@ async def run_noninteractive(
 
 
 def _validate_session_options(session: str | None, continue_last: bool) -> None:
-    """Reject conflicting session options."""
+    """Reject conflicting --session and --continue options."""
 
     if session and continue_last:
         raise typer.BadParameter("--session and --continue are mutually exclusive")