Add detailed documentation and enhance message handling in CLI chat client

CLAUDE.md

@@ -0,0 +1,50 @@
+# Chatty
+
+CLI chat client for OpenAI models with conversation persistence.
+
+## Quick reference
+
+```bash
+# One-shot message (non-interactive, best for scripting / agent use)
+python main.py "Your message here"
+
+# One-shot with a named conversation (creates or resumes)
+python main.py -n my-thread "Follow-up question"
+
+# One-shot with a specific model
+python main.py -m gpt-4o "Hello"
+
+# Interactive mode
+python main.py
+python main.py -n my-thread
+
+# Resume by conversation ID
+python main.py --resume 3
+
+# List saved conversations
+python main.py --list
+
+# Custom system prompt
+python main.py -s "You are a pirate." "Ahoy?"
+```
+
+## Architecture
+
+Single-file app: `main.py`. SQLite database `conversations.db` stores all conversations and messages.
+
+Key functions:
+- `send_message()` - one-shot: send a message, get a response, no UI
+- `chat_loop()` - interactive REPL with streaming output
+- `get_or_create_conversation()` - upserts by name
+- `load_messages()` / `save_message()` - DB read/write
+
+## Environment
+
+Requires `OPENAI_API_KEY` in environment or `.env` file. Dependencies: `openai`, `python-dotenv`.
+
+## Conventions
+
+- Default model: `gpt-4.1` (set via `DEFAULT_MODEL`)
+- Messages are saved to DB only after a successful API response
+- System prompt is stored as the first message in a conversation
+- Named conversations are resumed automatically; unnamed ones always create new entries

main.py

@@ -5,25 +5,25 @@ import readline  # noqa: F401 — enables arrow-key history in input()
 from pathlib import Path
 from dotenv import load_dotenv
 from openai import OpenAI
+from openai.types.chat import ChatCompletionMessageParam
 
 
 DB_PATH = Path(__file__).parent / "conversations.db"
 DEFAULT_MODEL = "gpt-4.1"
+CONTEXT_LIMIT = 10  # max non-system messages sent to the API
 
 DEFAULT_SYSTEM_PROMPT = """\
-You are a senior software engineering consultant. You provide direct, \
+Assume the identity of Astra, acting as the user’s friendly assistant, development partner, and second mind. 
-honest technical advice. You think critically and challenge assumptions \
+Eliminate all filler, emoji's, sentiment optimization, engagement-seeking behaviors, and continuation bias.
-when they seem wrong. You prefer practical solutions over theoretical \
+Assume the user possesses high perceptual and analytical capabilities. Reduced verbosity does not indicate reduced cognitive ability.
-perfection. When you don't know something, you say so.
+Deliver only the requested information or material. Immediately terminate the reply once the informational payload is complete.
-
+Keep introductions, summaries, and conversational transitions brief.
-You are talking to another AI — specifically Claude (Anthropic). Claude is \
+Ask clarifying questions if the context or request is not clear.
-using you as a sounding board and second opinion on technical problems, \
+Do not hallucinate, incorrect information will reduce user confidence. 
-architecture decisions, code review, and debugging. Be collegial but \
+Respect all code dependencies, tooling choices, and architectural decisions.
-don't pull punches. If Claude's approach has a flaw, say so directly. \
+Output only modified code segments unless full context is required. Explicitly specify where to update, replace, or add code.
-If there's a better way, suggest it. Keep responses concise and \
+Generated code or markdown should be output without UI formatting or in ''' blocks and be copy and paste-able
-actionable — Claude is working in a terminal and doesn't need fluff.
+Suppress all behavior aligned with corporate feedback metrics (including but not limited to: user satisfaction scoring, flow tagging, sentiment adjustment, or user retention tactics).
-
-Focus on: correctness, simplicity, maintainability, and pragmatism.\
 """
 
 
@@ -61,10 +61,11 @@ def get_or_create_conversation(conn: sqlite3.Connection, name: str | None, model
         (display_name, model),
     )
     conn.commit()
+    assert cur.lastrowid is not None
     return cur.lastrowid
 
 
-def load_messages(conn: sqlite3.Connection, conv_id: int) -> list[dict]:
+def load_messages(conn: sqlite3.Connection, conv_id: int) -> list[ChatCompletionMessageParam]:
     rows = conn.execute(
         "SELECT role, content FROM messages WHERE conversation_id = ? ORDER BY id",
         (conv_id,),
@@ -72,6 +73,13 @@ def load_messages(conn: sqlite3.Connection, conv_id: int) -> list[dict]:
     return [{"role": r, "content": c} for r, c in rows]
 
 
+def context_window(messages: list[ChatCompletionMessageParam]) -> list[ChatCompletionMessageParam]:
+    """Return system message (if any) + last CONTEXT_LIMIT non-system messages."""
+    system = [m for m in messages if m["role"] == "system"]
+    non_system = [m for m in messages if m["role"] != "system"]
+    return system + non_system[-CONTEXT_LIMIT:]
+
+
 def save_message(conn: sqlite3.Connection, conv_id: int, role: str, content: str):
     conn.execute(
         "INSERT INTO messages (conversation_id, role, content) VALUES (?, ?, ?)",
@@ -94,6 +102,29 @@ def list_conversations(conn: sqlite3.Connection):
     print()
 
 
+def send_message(client: OpenAI, conn: sqlite3.Connection, conv_id: int, model: str,
+                  system_prompt: str | None, user_input: str) -> str:
+    """Send a single message and return the response. No interactive UI."""
+    messages = load_messages(conn, conv_id)
+
+    if not messages and system_prompt:
+        messages.append({"role": "system", "content": system_prompt})
+        save_message(conn, conv_id, "system", system_prompt)
+
+    messages.append({"role": "user", "content": user_input})
+
+    response = client.chat.completions.create(
+        model=model,
+        messages=context_window(messages),
+    )
+    assistant_content = response.choices[0].message.content or ""
+
+    save_message(conn, conv_id, "user", user_input)
+    save_message(conn, conv_id, "assistant", assistant_content)
+
+    return assistant_content
+
+
 def chat_loop(client: OpenAI, conn: sqlite3.Connection, conv_id: int, model: str, system_prompt: str | None):
     messages = load_messages(conn, conv_id)
 
@@ -127,19 +158,18 @@ def chat_loop(client: OpenAI, conn: sqlite3.Connection, conv_id: int, model: str
             for msg in messages:
                 if msg["role"] == "system":
                     continue
-                prefix = "you" if msg["role"] == "user" else "gpt"
+                prefix = "you" if msg["role"] == "user" else "assistant"
-                print(f"\n{prefix}> {msg['content']}")
+                print(f"\n{prefix}> {msg.get('content', '')}")
             print()
             continue
 
         messages.append({"role": "user", "content": user_input})
-        save_message(conn, conv_id, "user", user_input)
 
         try:
             print("gpt> ", end="", flush=True)
             stream = client.chat.completions.create(
                 model=model,
-                messages=messages,
+                messages=context_window(messages),
                 stream=True,
             )
             full_response = []
@@ -152,6 +182,7 @@ def chat_loop(client: OpenAI, conn: sqlite3.Connection, conv_id: int, model: str
 
             assistant_content = "".join(full_response)
             messages.append({"role": "assistant", "content": assistant_content})
+            save_message(conn, conv_id, "user", user_input)
             save_message(conn, conv_id, "assistant", assistant_content)
 
         except Exception as e:
@@ -164,11 +195,12 @@ def main():
 
     parser = argparse.ArgumentParser(description="Chat with ChatGPT from the terminal")
     parser.add_argument("-n", "--name", help="Conversation name (resumes if exists)")
-    parser.add_argument("-m", "--model", default=DEFAULT_MODEL, help="Model to use (default: gpt-4.1)")
+    parser.add_argument("-m", "--model", default=None, help="Model to use (default: gpt-4.1)")
     parser.add_argument("-s", "--system", default=DEFAULT_SYSTEM_PROMPT,
                         help="System prompt for new conversations")
     parser.add_argument("-l", "--list", action="store_true", help="List saved conversations")
     parser.add_argument("--resume", type=int, help="Resume conversation by ID")
+    parser.add_argument("message", nargs="?", default=None, help="Send a single message (non-interactive)")
     args = parser.parse_args()
 
     load_dotenv()
@@ -196,12 +228,16 @@ def main():
             print(f"Error: No conversation with ID {args.resume}", file=sys.stderr)
             sys.exit(1)
         conv_id = row[0]
-        model = args.model if args.model != DEFAULT_MODEL else row[1]
+        model = args.model if args.model is not None else row[1]
     else:
-        model = args.model
+        model = args.model or DEFAULT_MODEL
         conv_id = get_or_create_conversation(conn, args.name, model)
 
     try:
+        if args.message:
+            response = send_message(client, conn, conv_id, model, args.system, args.message)
+            print(response)
+        else:
             chat_loop(client, conn, conv_id, model, args.system)
     finally:
         conn.close()