Plenty of new features

CLAUDE.md

@@ -55,7 +55,7 @@ The project uses a `src/smolagents/` layout. All source code is under `src/smola
 - `prompts/` - YAML prompt templates for different agent types
 
 ### Beyond Python (BP) Extensions
-- `bp_cli.py` - Interactive CLI (`bpsa`): REPL and one-shot modes with CodeAgent, slash commands, token tracking
+- `bp_cli.py` - Interactive CLI (`bpsa`): REPL and one-shot modes with CodeAgent, slash commands, token tracking, shell escapes (`!`, `!!`, `!!!`), `/alias`, `/redo`, auto-save
 - `bp_tools.py` - Extended tool library: file I/O, source code analysis, OS commands, multi-language support (24+ languages including Pascal, PHP, C++, Java, Go, Rust)
 - `bp_thinkers.py` - `Thinker` agent with multi-step reasoning (thoughts/plans/code sections)
 - `bp_executors.py` - `LocalExecExecutor` for direct Python execution via `exec()`

CLI.md

@@ -108,6 +108,14 @@ Use `prompt_toolkit` for:
 - Ctrl+C handling (cancel current input, not exit)
 - Autocomplete (slash commands)
 
+### Shell Escapes
+
+| Command | Description |
+|---------|-------------|
+| `!<command>` | Run an OS command directly (agent does not see the output) |
+| `!!<command>` | Run an OS command; output is appended to the next prompt sent to the agent |
+| `!!!<command>` | Run an OS command and immediately send the output to the agent for analysis |
+
 ### Step Display
 
 - Show intermediate code/thoughts as they happen (streaming)
@@ -126,12 +134,14 @@ Use `prompt_toolkit` for:
 | `/compression-max-uncompressed-steps <N>` | Change max_uncompressed_steps |
 | `/compression-model <model>` | Switch compression model |
 | `/exit` | Exit the REPL |
-| `/file <path>` | Load a file's content as the prompt |
 | `/help` | Show available commands and brief descriptions |
 | `/load-instructions` | Load agent instruction files into next prompt |
 | `/plan [on\|off\|N]` | Toggle or set planning interval (default: 22) |
 | `/pwd` | Show current working directory |
-| `/run <script.py>` | Execute a Python script |
+| `/repeat <N> <prompt>` | Run the same prompt N times, each on a fresh agent with current context |
+| `/repeat-prompt <N> <path>` | Run a prompt file N times, each on a fresh agent with current context |
+| `/run-prompt <path>` | Load a file's content as the prompt |
+| `/run-py <script.py>` | Execute a Python script |
 | `/save <filename>` | Save the last answer to a file |
 | `/save-step <N> <file>` | Save full content of step N to a file |
 | `/session-load <file>` | Load a session from a JSON file |
@@ -141,8 +151,8 @@ Use `prompt_toolkit` for:
 | `/show-stats` | Show session statistics (token usage, time) |
 | `/show-step <N>` | Show full content of a specific step |
 | `/show-steps` | Show one-line summary of all memory steps |
-| `/steps <N>` | Change max_steps for the agent |
-| `/tools` | List all loaded tools |
+| `/set-max-steps <N>` | Change max_steps for the agent |
+| `/show-tools` | List all loaded tools |
 | `/undo-steps [N]` | Remove last N steps from memory (default: 1) |
 | `/verbose` | Toggle verbose output |
 

README.md

@@ -66,7 +66,23 @@ $ bpsa --load-instructions          # Load CLAUDE.md, AGENTS.md, etc. at startup
 $ bpsa --browser                    # Enable Playwright browser integration
 ```
 
-The REPL supports command history, tab completion for slash commands, and multi-line input via Alt+Enter. Use `/session-save <file>` and `/session-load <file>` to persist and restore sessions across restarts.
+The REPL supports command history, tab completion for slash commands, and multi-line input via Alt+Enter. Use `/session-save <file>` and `/session-load <file>` to persist and restore sessions across restarts. You can also launch `ad-infinitum` from within the REPL via `!ad-infinitum ...`. Type `/help` to see all available commands.
+
+#### Shell commands from the REPL
+
+| Prefix | Description |
+|--------|-------------|
+| `!<command>` | Run an OS command directly (agent does not see the output) |
+| `!!<command>` | Run an OS command with streaming output; output is appended to the next prompt sent to the agent |
+| `!!!<command>` | Run an OS command and immediately send the output to the agent for analysis |
+
+#### Aliases
+
+Define command aliases with `/alias <name> <value>` (e.g., `/alias gs !!git status`). Aliases are saved to `~/.bpsa_aliases` and persist across sessions. Use `/alias` to list all and `/alias -d <name>` to delete.
+
+#### Auto-save
+
+Sessions are automatically saved every 5 turns to `~/.bpsa_autosave.json`. Configure the interval with the `BPSA_AUTOSAVE_INTERVAL` environment variable (set to 0 to disable).
 
 
 ## CLI (`ad-infinitum`)

src/smolagents/agents.py

@@ -687,19 +687,23 @@ def _finalize_step(self, memory_step: ActionStep | PlanningStep):
 
     def _handle_max_steps_reached(self, task: str) -> Any:
         action_step_start_time = time.time()
-        final_answer = self.provide_final_answer(task)
+        # BP: replaced forced LLM final answer with a static message to save tokens
+        # final_answer = self.provide_final_answer(task)
+        message = f"Max steps reached ({self.max_steps}/{self.max_steps}) - should I continue?"
         final_memory_step = ActionStep(
             step_number=self.step_number,
             error=AgentMaxStepsError("Reached max steps.", self.logger),
             timing=Timing(start_time=action_step_start_time, end_time=time.time()),
-            token_usage=final_answer.token_usage,
+            # token_usage=final_answer.token_usage,
             actionstep_id=self._next_actionstep_id,
         )
         self._next_actionstep_id += 1
-        final_memory_step.action_output = final_answer.content
+        # final_memory_step.action_output = final_answer.content
+        final_memory_step.action_output = message
         self._finalize_step(final_memory_step)
         self.memory.steps.append(final_memory_step)
-        return final_answer.content
+        # return final_answer.content
+        return message
 
     def _generate_planning_step(
         self, task, is_first_step: bool, step: int
@@ -1431,6 +1435,7 @@ def initialize_system_prompt(self) -> str:
                 "tools": self.tools,
                 "managed_agents": self.managed_agents,
                 "custom_instructions": self.instructions,
+                "model_id": getattr(self.model, 'model_id', 'unknown') or 'unknown',
             },
         )
         return system_prompt
@@ -1810,6 +1815,7 @@ def initialize_system_prompt(self) -> str:
                 "custom_instructions": self.instructions,
                 "code_block_opening_tag": self.code_block_tags[0],
                 "code_block_closing_tag": self.code_block_tags[1],
+                "model_id": getattr(self.model, 'model_id', 'unknown') or 'unknown',
             },
         )
         return system_prompt

src/smolagents/bp_ad_infinitum.py

@@ -29,7 +29,7 @@
     BPSA_PLAN_INTERVAL  - Planning interval (default: None = off)
     BPSA_MAX_STEPS      - Max steps per agent run (default: 200)
     BPSA_COOLDOWN       - Seconds to wait between cycles (default: 0)
-    BPSA_INJECT_FOLDER  - Inject directory tree (default: false, true = cwd, or a path)
+    BPSA_INJECT_FOLDER  - Inject directory tree (default: true = cwd, false = off, or a path)
 """
 
 import glob
@@ -177,19 +177,8 @@ def get_int_env(name: str, default: int) -> int:
 
 def inject_tree(folder: str) -> str:
     """Generate directory tree string to append to task prompts."""
-    from smolagents.bp_tools import list_directory_tree
-    tree = list_directory_tree(folder_path=folder, add_function_signatures=True)
-    return (
-        "\nThis is the result of list_directory_tree:\n<directory_tree>\n"
-        + tree
-        + "\n</directory_tree>\n"
-        "The contents of <directory_tree></directory_tree> is VERY important to you. "
-        "From <directory_tree></directory_tree>, you can get a general view/current state of the project:\n"
-        "* From the md files, if they exist, you can find the existing section titles "
-        "and have a general idea of the md file contents.\n"
-        "* For source code files, if they exist, you can find class and method names "
-        "so you can also develop a general idea of their contents.\n"
-    )
+    from smolagents.bp_tools import inject_tree as _inject_tree
+    return _inject_tree(folder)
 
 
 def run_script(task: TaskItem) -> subprocess.CompletedProcess:
@@ -368,9 +357,9 @@ def main():
     max_steps = get_int_env("BPSA_MAX_STEPS", 200)
     cooldown = get_int_env("BPSA_COOLDOWN", 0)
     tree_folder_raw = get_env("BPSA_INJECT_FOLDER")
-    if tree_folder_raw is None or tree_folder_raw.lower() == "false":
+    if tree_folder_raw is not None and tree_folder_raw.lower() == "false":
         tree_folder = None
-    elif tree_folder_raw.lower() == "true":
+    elif tree_folder_raw is None or tree_folder_raw.lower() == "true":
         tree_folder = os.getcwd()
     else:
         tree_folder = tree_folder_raw