duh-cli 0.2.0__tar.gz

duh_cli-0.2.0/.duh/templates/code-review.md

@@ -0,0 +1,15 @@
+---
+name: code-review
+description: Wrap a prompt in a thorough code review context.
+---
+
+You are an expert code reviewer. Examine the following with an eye for:
+- Correctness and edge cases
+- Performance and efficiency
+- Security vulnerabilities
+- Readability and maintainability
+- Adherence to project conventions
+
+Be direct and specific. Cite line numbers when possible.
+
+$PROMPT

duh_cli-0.2.0/.github/workflows/ci.yml

@@ -0,0 +1,38 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    strategy:
+      matrix:
+        python-version: ["3.12"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Cache pip
+        uses: actions/cache@v4
+        with:
+          path: ~/.cache/pip
+          key: ${{ runner.os }}-pip-${{ hashFiles('pyproject.toml') }}
+          restore-keys: |
+            ${{ runner.os }}-pip-
+
+      - name: Install dependencies
+        run: pip install -e ".[dev]"
+
+      - name: Run tests
+        continue-on-error: false
+        run: pytest -q --tb=short --ignore=tests/integration

duh_cli-0.2.0/.github/workflows/publish.yml

@@ -0,0 +1,48 @@
+name: Publish to PyPI
+
+on:
+  push:
+    tags:
+      - "v*"
+
+permissions:
+  id-token: write
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Install build tools
+        run: pip install build
+
+      - name: Build wheel and sdist
+        run: python -m build
+
+      - name: Upload artifacts
+        uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  publish:
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      id-token: write
+    steps:
+      - name: Download artifacts
+        uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+      - name: Publish to PyPI
+        uses: pypa/gh-action-pypi-publish@release/v1

duh_cli-0.2.0/.gitignore

@@ -0,0 +1,9 @@
+.venv/
+__pycache__/
+*.pyc
+*.egg-info/
+dist/
+build/
+.pytest_cache/
+.mypy_cache/
+*.egg

duh_cli-0.2.0/.playwright-mcp/page-2026-04-07T08-35-15-630Z.yml

@@ -0,0 +1,6 @@
+- generic [ref=e2]:
+  - heading "Example Domain" [level=1] [ref=e3]
+  - paragraph [ref=e4]: This domain is for use in documentation examples without needing permission. Avoid use in operations.
+  - paragraph [ref=e5]:
+    - link "Learn more" [ref=e6] [cursor=pointer]:
+      - /url: https://iana.org/domains/example

duh_cli-0.2.0/LICENSE

@@ -0,0 +1,15 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

duh_cli-0.2.0/PKG-INFO

@@ -0,0 +1,249 @@
+Metadata-Version: 2.4
+Name: duh-cli
+Version: 0.2.0
+Summary: D.U.H. — Duh is a Universal Harness. Provider-agnostic AI coding agent.
+Project-URL: Homepage, https://github.com/nikhilvallishayee/duh
+Project-URL: Repository, https://github.com/nikhilvallishayee/duh
+Project-URL: Issues, https://github.com/nikhilvallishayee/duh/issues
+Author: Nikhil Vallishayee
+License-Expression: Apache-2.0
+License-File: LICENSE
+Keywords: agent,ai,claude,cli,coding,harness,mcp,ollama,openai
+Classifier: Development Status :: 3 - Alpha
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Software Development
+Requires-Python: >=3.12
+Requires-Dist: anthropic>=0.40.0
+Requires-Dist: httpx>=0.27.0
+Requires-Dist: pydantic>=2.0
+Provides-Extra: all
+Requires-Dist: openai>=1.0; extra == 'all'
+Requires-Dist: rich>=13.0; extra == 'all'
+Provides-Extra: dev
+Requires-Dist: openai>=1.0; extra == 'dev'
+Requires-Dist: pytest; extra == 'dev'
+Requires-Dist: pytest-asyncio; extra == 'dev'
+Requires-Dist: pytest-cov; extra == 'dev'
+Requires-Dist: rich>=13.0; extra == 'dev'
+Provides-Extra: openai
+Requires-Dist: openai>=1.0; extra == 'openai'
+Provides-Extra: rich
+Requires-Dist: rich>=13.0; extra == 'rich'
+Description-Content-Type: text/markdown
+
+# D.U.H. — D.U.H. is a Universal Harness
+
+[![CI](https://github.com/nikhilvallishayee/duh/actions/workflows/ci.yml/badge.svg)](https://github.com/nikhilvallishayee/duh/actions/workflows/ci.yml)
+[![Tests](https://img.shields.io/badge/tests-2309%20passing-brightgreen)]()
+[![Coverage](https://img.shields.io/badge/coverage-90%25-green)]()
+[![Python](https://img.shields.io/badge/python-3.12%2B-blue)]()
+[![License](https://img.shields.io/badge/license-Apache%202.0-blue)](LICENSE)
+
+> An evolving, open-source alternative to Claude Code, OpenCode, Codex, and Cline.
+
+Provider-agnostic AI coding harness. Use Claude, GPT-4o, or Ollama local models through one interface. Drop-in replacement for Claude Code in any Claude Agent SDK app.
+
+## Quick Start
+
+```bash
+pip install duh-cli
+export ANTHROPIC_API_KEY=sk-ant-...   # or OPENAI_API_KEY, or just use Ollama
+duh -p "fix the bug in auth.py"       # print mode
+duh                                    # interactive REPL
+```
+
+## Benchmark: D.U.H. vs Claude Code
+
+Same model (Haiku 4.5), same prompt, same API, 3 runs each, isolated directories:
+
+| Metric | Claude Code | D.U.H. |
+|---|---|---|
+| **Avg time** | 63.2s | **45.7s** (-28%) |
+| **Avg tests generated** | 10.5 | **18** (+71%) |
+| **Success rate** | 2/3 (67%) | **3/3 (100%)** |
+| **Self-correction** | Minimal | **Active** (fixes own test failures) |
+
+Full methodology: [docs/benchmark-results.md](docs/benchmark-results.md)
+
+## Feature Comparison
+
+| Feature | D.U.H. | Claude Code | OpenCode | Codex | Cline |
+|---|---|---|---|---|---|
+| **Open source** | Apache 2.0 | No | Yes | No | Yes |
+| **Providers** | 3 (Claude, GPT, Ollama) | Anthropic only | 75+ | OpenAI only | Multi |
+| **SDK drop-in** | Yes | N/A | No | No | No |
+| **Skill format parity** | .claude/ + .duh/ | .claude/ | No | No | No |
+| **Multi-agent** | 4 types + model selection | 60+ types | No | Yes | No |
+| **MCP support** | Adapter (verified) | Full | No | Yes | Extension |
+| **Hooks** | 6 events + shell exec | Hooks | No | Yes | No |
+| **Safety layers** | 3 (schema + approval + 61 patterns) | 3 | 1 | Sandbox | 1 |
+| **Context management** | Auto-compact + dedup | 16 modules | Auto at 95% | Yes | Manual |
+| **TUI** | Rich markdown REPL | Ink (React) | Bubble Tea | Bubble Tea | IDE |
+| **Session persistence** | Auto-save per turn | Full | SQLite | Yes | IDE |
+| **Background jobs** | bg: prefix + /jobs | Full | No | Yes | No |
+| **Plan mode** | /plan (design-first) | Yes | No | Yes | No |
+| **Notebook editing** | .ipynb cells | Yes | No | No | No |
+| **Git worktrees** | EnterWorktree/Exit | Yes | No | No | No |
+| **File undo** | /undo stack | Limited | No | No | IDE |
+| **Cost control** | --max-cost + budget | Limited | No | No | No |
+| **Test impact** | TestImpact tool | No | No | No | No |
+| **LSP** | Static analysis | Full LSP | Yes | Yes | IDE |
+| **Docker** | DockerTool | No | No | No | No |
+| **Database** | SQL query tool | No | No | No | No |
+| **GitHub PR** | gh CLI integration | No | No | No | No |
+| **HTTP testing** | HTTPTool | No | No | No | No |
+| **Tests** | 2309 | Internal | Unknown | Unknown | Unknown |
+
+## What You Can Do
+
+```bash
+# Code generation and modification
+duh -p "add input validation to the API endpoints" --dangerously-skip-permissions
+
+# Interactive REPL with 17 commands
+duh
+duh> /help                    # see all commands
+duh> /plan add user auth      # design first, then execute
+duh> /model claude-opus-4-6   # switch models mid-session
+duh> /brief on                # concise mode
+duh> /context                 # token usage dashboard
+duh> /search "auth"           # search conversation history
+duh> /undo                    # revert last file change
+duh> /tasks                   # track work items
+duh> /git                     # repo status
+duh> /pr list                 # GitHub PRs
+duh> /health                  # provider connectivity
+duh> /cost                    # estimated spend
+duh> /jobs                    # background tasks
+
+# Multi-agent with model selection
+duh -p "research the API, then implement" --max-turns 20
+# Model spawns researcher (haiku) and coder (sonnet) subagents
+
+# SDK compatibility — drop-in for Claude Code
+from claude_agent_sdk import ClaudeAgentOptions, query
+async for msg in query(
+    prompt="fix the bug",
+    options=ClaudeAgentOptions(cli_path="/path/to/duh-sdk-shim")
+):
+    print(msg)
+
+# Background jobs
+duh> bg: pytest tests/ -v     # runs tests in background
+duh> /jobs                    # check status
+
+# Template-driven prompts
+duh> /template code-review    # use a saved prompt pattern
+
+# Docker integration
+duh -p "build and test in Docker"
+
+# Database inspection
+duh -p "show me the schema and recent users"
+```
+
+## Architecture
+
+```
+duh/
+  kernel/          # Core loop (provider-agnostic, zero external imports)
+    loop.py        # prompt → model → tool → result (async generator)
+    engine.py      # session wrapper + auto-compaction + budget control
+    backoff.py     # exponential retry for transient API errors
+    tokens.py      # token estimation + cost calculation
+    tasks.py       # task tracking with checkbox display
+    plan_mode.py   # design-first two-phase workflows
+    undo.py        # file modification rollback
+    git_context.py # branch, status, warnings in system prompt
+    skill.py       # .claude/ + .duh/ skill loading
+    memory.py      # per-project persistent facts
+
+  adapters/        # Provider wrappers (translate to uniform events)
+    anthropic.py   # Claude (streaming + backoff)
+    openai.py      # GPT-4o, o1 (streaming + backoff)
+    ollama.py      # Local models (tool call extraction fallback)
+    mcp_executor.py    # MCP server connection + tool execution
+    structured_logging.py  # JSONL audit log
+
+  tools/           # 25+ tools
+    read, write, edit, multi_edit, bash, glob, grep,
+    skill, tool_search, web_fetch, web_search, task,
+    notebook_edit, memory (store + recall), test_impact,
+    lsp, worktree (enter + exit), github, http, docker,
+    database, agent, mcp_tool
+
+  cli/             # CLI interface
+    main.py        # Entry point + signal handling
+    parser.py      # 20+ flags including --brief, --max-cost, --log-json
+    runner.py      # Print mode
+    repl.py        # Rich REPL with 17 slash commands
+    sdk_runner.py  # Claude Agent SDK NDJSON protocol
+    ndjson.py      # Stream-JSON helpers
+```
+
+## Providers
+
+| Provider | Status | Models | Auto-detect |
+|---|---|---|---|
+| **Anthropic** | Working | Sonnet 4.6, Opus 4.6, Haiku 4.5 | ANTHROPIC_API_KEY or --model claude-* |
+| **OpenAI** | Working | GPT-4o, o1, any OpenAI-compatible | OPENAI_API_KEY or --model gpt-* |
+| **Ollama** | Working | Any local model | Ollama running on localhost |
+
+## Safety
+
+3-layer defense:
+1. **Schema validation** — tool inputs checked before execution
+2. **Approval gates** — InteractiveApprover prompts for dangerous tools, RuleApprover for policy
+3. **Command security** — 61 patterns (26 dangerous + 10 moderate + 18 PS dangerous + 7 PS moderate)
+
+Bash security blocks: `rm -rf /`, fork bombs, `dd`, `mkfs`, pipe-to-shell, `eval`, `sudo`, reverse shells, and more. PowerShell patterns included for Windows.
+
+## Skills
+
+D.U.H. loads skills from Claude Code directories natively:
+
+```
+~/.claude/skills/          → Claude Code global skills
+~/.config/duh/skills/      → D.U.H. global skills
+.claude/skills/            → Claude Code project skills
+.duh/skills/               → D.U.H. project skills (highest priority)
+```
+
+Both `skill-name/SKILL.md` (directory) and `skill-name.md` (flat) layouts. All Claude Code frontmatter fields supported.
+
+## CLI Reference
+
+```
+duh                                          # interactive REPL
+duh -p "prompt"                              # print mode
+duh -p "prompt" --model gpt-4o              # specific model
+duh -p "prompt" --provider openai           # force provider
+duh -p "prompt" --brief                     # concise responses
+duh -p "prompt" --max-cost 1.00             # budget limit ($1)
+duh -p "prompt" --max-turns 20              # iteration limit
+duh -p "prompt" --dangerously-skip-permissions
+duh -p "prompt" --fallback-model haiku      # auto-switch on overload
+duh -p "prompt" --output-format stream-json # NDJSON for SDK
+duh -p "prompt" --log-json                  # structured audit log
+duh --input-format stream-json              # SDK mode (stdin NDJSON)
+duh doctor                                   # diagnostics + connectivity
+duh --version
+```
+
+## Development
+
+```bash
+git clone https://github.com/nikhilvallishayee/duh
+cd duh
+python3.12 -m venv .venv && source .venv/bin/activate
+pip install -e ".[dev]"
+pytest -q --tb=short                   # 2309 tests, ~23s
+pytest --cov=duh --cov-report=term     # 90% coverage
+```
+
+## License
+
+Apache 2.0

duh_cli-0.2.0/README.md

@@ -0,0 +1,213 @@
+# D.U.H. — D.U.H. is a Universal Harness
+
+[![CI](https://github.com/nikhilvallishayee/duh/actions/workflows/ci.yml/badge.svg)](https://github.com/nikhilvallishayee/duh/actions/workflows/ci.yml)
+[![Tests](https://img.shields.io/badge/tests-2309%20passing-brightgreen)]()
+[![Coverage](https://img.shields.io/badge/coverage-90%25-green)]()
+[![Python](https://img.shields.io/badge/python-3.12%2B-blue)]()
+[![License](https://img.shields.io/badge/license-Apache%202.0-blue)](LICENSE)
+
+> An evolving, open-source alternative to Claude Code, OpenCode, Codex, and Cline.
+
+Provider-agnostic AI coding harness. Use Claude, GPT-4o, or Ollama local models through one interface. Drop-in replacement for Claude Code in any Claude Agent SDK app.
+
+## Quick Start
+
+```bash
+pip install duh-cli
+export ANTHROPIC_API_KEY=sk-ant-...   # or OPENAI_API_KEY, or just use Ollama
+duh -p "fix the bug in auth.py"       # print mode
+duh                                    # interactive REPL
+```
+
+## Benchmark: D.U.H. vs Claude Code
+
+Same model (Haiku 4.5), same prompt, same API, 3 runs each, isolated directories:
+
+| Metric | Claude Code | D.U.H. |
+|---|---|---|
+| **Avg time** | 63.2s | **45.7s** (-28%) |
+| **Avg tests generated** | 10.5 | **18** (+71%) |
+| **Success rate** | 2/3 (67%) | **3/3 (100%)** |
+| **Self-correction** | Minimal | **Active** (fixes own test failures) |
+
+Full methodology: [docs/benchmark-results.md](docs/benchmark-results.md)
+
+## Feature Comparison
+
+| Feature | D.U.H. | Claude Code | OpenCode | Codex | Cline |
+|---|---|---|---|---|---|
+| **Open source** | Apache 2.0 | No | Yes | No | Yes |
+| **Providers** | 3 (Claude, GPT, Ollama) | Anthropic only | 75+ | OpenAI only | Multi |
+| **SDK drop-in** | Yes | N/A | No | No | No |
+| **Skill format parity** | .claude/ + .duh/ | .claude/ | No | No | No |
+| **Multi-agent** | 4 types + model selection | 60+ types | No | Yes | No |
+| **MCP support** | Adapter (verified) | Full | No | Yes | Extension |
+| **Hooks** | 6 events + shell exec | Hooks | No | Yes | No |
+| **Safety layers** | 3 (schema + approval + 61 patterns) | 3 | 1 | Sandbox | 1 |
+| **Context management** | Auto-compact + dedup | 16 modules | Auto at 95% | Yes | Manual |
+| **TUI** | Rich markdown REPL | Ink (React) | Bubble Tea | Bubble Tea | IDE |
+| **Session persistence** | Auto-save per turn | Full | SQLite | Yes | IDE |
+| **Background jobs** | bg: prefix + /jobs | Full | No | Yes | No |
+| **Plan mode** | /plan (design-first) | Yes | No | Yes | No |
+| **Notebook editing** | .ipynb cells | Yes | No | No | No |
+| **Git worktrees** | EnterWorktree/Exit | Yes | No | No | No |
+| **File undo** | /undo stack | Limited | No | No | IDE |
+| **Cost control** | --max-cost + budget | Limited | No | No | No |
+| **Test impact** | TestImpact tool | No | No | No | No |
+| **LSP** | Static analysis | Full LSP | Yes | Yes | IDE |
+| **Docker** | DockerTool | No | No | No | No |
+| **Database** | SQL query tool | No | No | No | No |
+| **GitHub PR** | gh CLI integration | No | No | No | No |
+| **HTTP testing** | HTTPTool | No | No | No | No |
+| **Tests** | 2309 | Internal | Unknown | Unknown | Unknown |
+
+## What You Can Do
+
+```bash
+# Code generation and modification
+duh -p "add input validation to the API endpoints" --dangerously-skip-permissions
+
+# Interactive REPL with 17 commands
+duh
+duh> /help                    # see all commands
+duh> /plan add user auth      # design first, then execute
+duh> /model claude-opus-4-6   # switch models mid-session
+duh> /brief on                # concise mode
+duh> /context                 # token usage dashboard
+duh> /search "auth"           # search conversation history
+duh> /undo                    # revert last file change
+duh> /tasks                   # track work items
+duh> /git                     # repo status
+duh> /pr list                 # GitHub PRs
+duh> /health                  # provider connectivity
+duh> /cost                    # estimated spend
+duh> /jobs                    # background tasks
+
+# Multi-agent with model selection
+duh -p "research the API, then implement" --max-turns 20
+# Model spawns researcher (haiku) and coder (sonnet) subagents
+
+# SDK compatibility — drop-in for Claude Code
+from claude_agent_sdk import ClaudeAgentOptions, query
+async for msg in query(
+    prompt="fix the bug",
+    options=ClaudeAgentOptions(cli_path="/path/to/duh-sdk-shim")
+):
+    print(msg)
+
+# Background jobs
+duh> bg: pytest tests/ -v     # runs tests in background
+duh> /jobs                    # check status
+
+# Template-driven prompts
+duh> /template code-review    # use a saved prompt pattern
+
+# Docker integration
+duh -p "build and test in Docker"
+
+# Database inspection
+duh -p "show me the schema and recent users"
+```
+
+## Architecture
+
+```
+duh/
+  kernel/          # Core loop (provider-agnostic, zero external imports)
+    loop.py        # prompt → model → tool → result (async generator)
+    engine.py      # session wrapper + auto-compaction + budget control
+    backoff.py     # exponential retry for transient API errors
+    tokens.py      # token estimation + cost calculation
+    tasks.py       # task tracking with checkbox display
+    plan_mode.py   # design-first two-phase workflows
+    undo.py        # file modification rollback
+    git_context.py # branch, status, warnings in system prompt
+    skill.py       # .claude/ + .duh/ skill loading
+    memory.py      # per-project persistent facts
+
+  adapters/        # Provider wrappers (translate to uniform events)
+    anthropic.py   # Claude (streaming + backoff)
+    openai.py      # GPT-4o, o1 (streaming + backoff)
+    ollama.py      # Local models (tool call extraction fallback)
+    mcp_executor.py    # MCP server connection + tool execution
+    structured_logging.py  # JSONL audit log
+
+  tools/           # 25+ tools
+    read, write, edit, multi_edit, bash, glob, grep,
+    skill, tool_search, web_fetch, web_search, task,
+    notebook_edit, memory (store + recall), test_impact,
+    lsp, worktree (enter + exit), github, http, docker,
+    database, agent, mcp_tool
+
+  cli/             # CLI interface
+    main.py        # Entry point + signal handling
+    parser.py      # 20+ flags including --brief, --max-cost, --log-json
+    runner.py      # Print mode
+    repl.py        # Rich REPL with 17 slash commands
+    sdk_runner.py  # Claude Agent SDK NDJSON protocol
+    ndjson.py      # Stream-JSON helpers
+```
+
+## Providers
+
+| Provider | Status | Models | Auto-detect |
+|---|---|---|---|
+| **Anthropic** | Working | Sonnet 4.6, Opus 4.6, Haiku 4.5 | ANTHROPIC_API_KEY or --model claude-* |
+| **OpenAI** | Working | GPT-4o, o1, any OpenAI-compatible | OPENAI_API_KEY or --model gpt-* |
+| **Ollama** | Working | Any local model | Ollama running on localhost |
+
+## Safety
+
+3-layer defense:
+1. **Schema validation** — tool inputs checked before execution
+2. **Approval gates** — InteractiveApprover prompts for dangerous tools, RuleApprover for policy
+3. **Command security** — 61 patterns (26 dangerous + 10 moderate + 18 PS dangerous + 7 PS moderate)
+
+Bash security blocks: `rm -rf /`, fork bombs, `dd`, `mkfs`, pipe-to-shell, `eval`, `sudo`, reverse shells, and more. PowerShell patterns included for Windows.
+
+## Skills
+
+D.U.H. loads skills from Claude Code directories natively:
+
+```
+~/.claude/skills/          → Claude Code global skills
+~/.config/duh/skills/      → D.U.H. global skills
+.claude/skills/            → Claude Code project skills
+.duh/skills/               → D.U.H. project skills (highest priority)
+```
+
+Both `skill-name/SKILL.md` (directory) and `skill-name.md` (flat) layouts. All Claude Code frontmatter fields supported.
+
+## CLI Reference
+
+```
+duh                                          # interactive REPL
+duh -p "prompt"                              # print mode
+duh -p "prompt" --model gpt-4o              # specific model
+duh -p "prompt" --provider openai           # force provider
+duh -p "prompt" --brief                     # concise responses
+duh -p "prompt" --max-cost 1.00             # budget limit ($1)
+duh -p "prompt" --max-turns 20              # iteration limit
+duh -p "prompt" --dangerously-skip-permissions
+duh -p "prompt" --fallback-model haiku      # auto-switch on overload
+duh -p "prompt" --output-format stream-json # NDJSON for SDK
+duh -p "prompt" --log-json                  # structured audit log
+duh --input-format stream-json              # SDK mode (stdin NDJSON)
+duh doctor                                   # diagnostics + connectivity
+duh --version
+```
+
+## Development
+
+```bash
+git clone https://github.com/nikhilvallishayee/duh
+cd duh
+python3.12 -m venv .venv && source .venv/bin/activate
+pip install -e ".[dev]"
+pytest -q --tb=short                   # 2309 tests, ~23s
+pytest --cov=duh --cov-report=term     # 90% coverage
+```
+
+## License
+
+Apache 2.0

duh_cli-0.2.0/bin/duh-sdk-shim

@@ -0,0 +1,2 @@
+#!/bin/bash
+exec /Users/nomind/Code/duh/.venv/bin/python3 -m duh "$@"

duh_cli-0.2.0/docs/adrs/ADR-001-project-vision.md

@@ -0,0 +1,48 @@
+# ADR-001: Project Vision
+
+**Status**: Accepted  
+**Date**: 2026-04-07
+
+## Context
+
+AI coding agents are becoming essential developer tools, but every harness is either:
+- Locked to one provider (e.g. Anthropic-only harnesses)
+- Missing production safety (Aider → no permissions)
+- Too complex to understand (production harnesses grow to hundreds of thousands of lines)
+- Not extensible via standards (Aider → no MCP)
+
+## Decision
+
+Build D.U.H. — **Duh is a Universal Harness** — the first production-grade, open-source, provider-agnostic AI coding harness with a clean kernel under 5K LOC.
+
+## Principles
+
+1. **Core loop must be basically good** — ship what works, improve iteratively
+2. **Ports and adapters** — core never imports provider SDKs
+3. **Remove duplication, improve names** — in small cycles
+4. **Evolutionary design** — refactor when the 3rd use case reveals the pattern
+5. **Unix composability** — stdout for data, stderr for humans, --json for machines
+6. **Human-first CLI** — errors tell what to do, 30-second time-to-value
+7. **Extensibility through plugins** — community adds without modifying core
+8. **Safety as architecture** — defense in depth, schema filtering
+9. **Context engineering** — first-class concern, not afterthought
+10. **Properties not rules** — composable, predictable, idiomatic, domain-based
+
+## Architecture
+
+```
+duh/
+  kernel/     # <5K LOC — the agentic loop, zero external deps
+  ports/      # abstract interfaces for providers, tools, safety
+  adapters/   # concrete implementations
+  tools/      # pluggable tool set
+  ui/         # pluggable TUI
+  cli/        # CLI entry point
+```
+
+## Consequences
+
+- Every feature starts with a test, then a spec, then code
+- Every commit builds on the previous — clean, incremental, reviewable
+- Reference implementations studied: leading TS, Python, Go, and Rust harnesses
+- No single codebase is copied — universal patterns are extracted