Manual Slop

A GUI orchestrator for local LLM-driven coding sessions. Manual Slop bridges high-latency AI reasoning with a low-latency ImGui render loop via a thread-safe asynchronous pipeline, ensuring every AI-generated payload passes through a human-auditable gate before execution.

Tech Stack: Python 3.11+, Dear PyGui / ImGui, FastAPI, Uvicorn Providers: Gemini API, Anthropic API, DeepSeek, Gemini CLI (headless) Platform: Windows (PowerShell) — single developer, local use

---

Architecture at a Glance

Four thread domains operate concurrently: the ImGui main loop, an asyncio worker for AI calls, a HookServer (HTTP on :8999) for external automation, and transient threads for model fetching. Background threads never write GUI state directly — they serialize task dicts into lock-guarded lists that the main thread drains once per frame (details).

The Execution Clutch suspends the AI execution thread on a threading.Condition when a destructive action (PowerShell script, sub-agent spawn) is requested. The GUI renders a modal where the user can read, edit, or reject the payload. On approval, the condition is signaled and execution resumes (details).

The MMA (Multi-Model Agent) system decomposes epics into tracks, tracks into DAG-ordered tickets, and executes each ticket with a stateless Tier 3 worker that starts from ai_client.reset_session() — no conversational bleed between tickets (details).

---

Documentation

Guide	Scope
Architecture	Threading model, event system, AI client multi-provider architecture, HITL mechanism, comms logging
Tools & IPC	MCP Bridge security model, all 26 native tools, Hook API endpoints, ApiHookClient reference, shell runner
MMA Orchestration	4-tier hierarchy, Ticket/Track data structures, DAG engine, ConductorEngine execution loop, worker lifecycle
Simulations	live_gui fixture, Puppeteer pattern, mock provider, visual verification patterns, ASTParser / summarizer

---

Module Map

File	Lines	Role
gui_2.py	~3080	Primary ImGui interface — App class, frame-sync, HITL dialogs
ai_client.py	~1800	Multi-provider LLM abstraction (Gemini, Anthropic, DeepSeek, Gemini CLI)
mcp_client.py	~870	26 MCP tools with filesystem sandboxing and tool dispatch
api_hooks.py	~330	HookServer — REST API for external automation on :8999
api_hook_client.py	~245	Python client for the Hook API (used by tests and external tooling)
multi_agent_conductor.py	~250	ConductorEngine — Tier 2 orchestration loop with DAG execution
conductor_tech_lead.py	~100	Tier 2 ticket generation from track briefs
dag_engine.py	~100	TrackDAG (dependency graph) + ExecutionEngine (tick-based state machine)
models.py	~100	Ticket, Track, WorkerContext dataclasses
events.py	~89	EventEmitter, AsyncEventQueue, UserRequestEvent
project_manager.py	~300	TOML config persistence, discussion management, track state
session_logger.py	~200	JSON-L + markdown audit trails (comms, tools, CLI, hooks)
shell_runner.py	~100	PowerShell execution with timeout, env config, QA callback
file_cache.py	~150	ASTParser (tree-sitter) — skeleton and curated views
summarize.py	~120	Heuristic file summaries (imports, classes, functions)
outline_tool.py	~80	Hierarchical code outline via stdlib ast

---

Setup

Prerequisites

• Python 3.11+
• uv for package management

Installation

git clone <repo>
cd manual_slop
uv sync

Credentials

Configure in credentials.toml:

[gemini]
api_key = "YOUR_KEY"

[anthropic]
api_key = "YOUR_KEY"

[deepseek]
api_key = "YOUR_KEY"

Running

uv run gui_2.py                        # Normal mode
uv run gui_2.py --enable-test-hooks    # With Hook API on :8999

Running Tests

uv run pytest tests/ -v

---

Project Configuration

Projects are stored as <name>.toml files. The discussion history is split into a sibling <name>_history.toml to keep the main config lean.

[project]
name = "my_project"
git_dir = "./my_repo"
system_prompt = ""

[files]
base_dir = "./my_repo"
paths = ["src/**/*.py", "README.md"]

[screenshots]
base_dir = "./my_repo"
paths = []

[output]
output_dir = "./md_gen"

[gemini_cli]
binary_path = "gemini"

[agent.tools]
run_powershell = true
read_file = true
# ... 26 tool flags