odd-org 0.1.0__tar.gz

odd_org-0.1.0/.github/dependabot.yml

@@ -0,0 +1,13 @@
+version: 2
+updates:
+  - package-ecosystem: "pip"
+    directory: "/"
+    schedule:
+      interval: "weekly"
+    open-pull-requests-limit: 5
+
+  - package-ecosystem: "github-actions"
+    directory: "/"
+    schedule:
+      interval: "weekly"
+    open-pull-requests-limit: 5

odd_org-0.1.0/.github/workflows/ci.yml

@@ -0,0 +1,64 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.11", "3.12", "3.13"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -r requirements.lock
+          pip install -e . --no-deps
+          pip install pytest pytest-cov
+
+      - name: Run tests with coverage
+        run: pytest --cov=odd --cov-report=term-missing --cov-report=xml
+
+      - name: Upload coverage to Codecov
+        if: matrix.python-version == '3.13'
+        uses: codecov/codecov-action@v4
+        with:
+          files: coverage.xml
+          fail_ci_if_error: false
+        env:
+          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
+
+  security:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.13"
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -r requirements.lock
+          pip install -e . --no-deps
+          pip install pip-audit bandit
+
+      - name: pip audit - CVE scanning
+        run: pip-audit
+
+      - name: Bandit - Python SAST
+        run: bandit -r src/odd/ -c pyproject.toml

odd_org-0.1.0/.gitignore

@@ -0,0 +1,12 @@
+__pycache__/
+*.py[cod]
+*$py.class
+*.egg-info/
+dist/
+build/
+.eggs/
+*.egg
+.venv/
+venv/
+.env
+.odd/

odd_org-0.1.0/CLAUDE.md

@@ -0,0 +1,38 @@
+# ODD - Organization Driven Development
+
+## What is this?
+A Python CLI tool that lets you be the CEO of an AI software company. Agents organized like a real company (Lead Dev, QA, Recruiter, Secretary, Consultants) collaborate through agile processes to build software.
+
+## Architecture
+- `src/odd/` - Main package
+  - `cli.py` - Click CLI entry point (`odd init`, `odd sprint`, `odd standup`, `odd team`, `odd hire`, `odd status`)
+  - `agent.py` - Agent invocation engine. Each agent is a strictly isolated API call via a Provider.
+  - `provider.py` - Provider protocol (the abstraction boundary between ODD and LLM APIs)
+  - `providers/` - Provider implementations (anthropic.py is the default)
+  - `models.py` - Pydantic data models (Persona, Sprint, ProjectConfig, etc.)
+  - `roles.py` - Core team definitions + recruiter pipeline (JD → resume)
+  - `sprint.py` - Sprint lifecycle engine (plan → QA writes tests → implement → demo)
+  - `state.py` - `.odd/` directory state management
+  - `tools.py` - File/shell tools in neutral format; providers wrap for their wire format
+
+## Key Design Principles
+- **Strict agent isolation**: Each agent is a separate API call with its own system prompt built from persona files. NEVER share conversation history across personas.
+- **Claude-default**: Anthropic is the default, best-tested provider. Other providers can plug in via the Provider protocol.
+- **TDD as quality gate**: QA writes tests first, agents implement to pass them.
+- **Scope-boxed sprints**: Sprint boundaries are CEO decision points, not time-based.
+
+## Commands
+```
+pip install -e .
+odd init          # Boot the org, talk to lead dev
+odd sprint        # Start a sprint
+odd standup       # Run a team standup
+odd team          # Show team roster
+odd hire "desc"   # Hire a consultant
+odd status        # Project status
+```
+
+## Testing
+```
+pytest
+```

odd_org-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Eric Smith
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

odd_org-0.1.0/PKG-INFO

@@ -0,0 +1,158 @@
+Metadata-Version: 2.4
+Name: odd-org
+Version: 0.1.0
+Summary: Organization Driven Development - AI agents organized like a real software company
+Project-URL: Repository, https://github.com/ThePoetCoder/odd
+Project-URL: Issues, https://github.com/ThePoetCoder/odd/issues
+License-Expression: MIT
+License-File: LICENSE
+Keywords: agents,ai,anthropic,claude,cli,development,organization
+Classifier: Development Status :: 3 - Alpha
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Software Development
+Classifier: Topic :: Software Development :: Testing
+Requires-Python: >=3.11
+Requires-Dist: anthropic>=0.40.0
+Requires-Dist: click>=8.1.0
+Requires-Dist: pydantic>=2.0.0
+Requires-Dist: rich>=13.0.0
+Provides-Extra: openai-compat
+Requires-Dist: requests>=2.28.0; extra == 'openai-compat'
+Description-Content-Type: text/markdown
+
+# ODD - Organization Driven Development; an Executable Metaphor
+
+**We live in odd times**
+
+## The Idea
+
+You don't write code. You don't write specs. You run a company.
+
+ODD organizes AI agents like a real software team:
+* Lead Dev
+* QA
+* Recruiter
+* Secretary
+* CFO
+* Specialist Consultants
+
+They all collaborate through TDD and agile processes to build software. You provide vision and make decisions at sprint boundaries. They do everything else.
+
+This is the next step up the abstraction ladder:
+
+1. **You write code** to solve your problem
+2. **You write specs** for an agent to code
+3. **You direct a company of agents** and they spec, build, test, and ship.
+
+## Why Organizational Patterns?
+
+Sprints, standups, TDD, code review, retrospectives - these aren't just corporate rituals. They're coordination algorithms refined over decades of real software teams solving real problems. They evolved to handle communication overhead, context switching, uncertainty, and quality control.
+
+AI agents have the same coordination problems. ODD applies the patterns that already work:
+
+- **Strict agent isolation** - each team member is a separate API call with their own persona. No shared memory, no context bleed between roles. The same reason you don't want your QA engineer thinking like your developer.
+- **TDD as quality gate** - QA writes tests *before* implementation. Tests passing is a more objective quality signal than code review alone, but you can add code review too.
+- **Scope-boxed sprints** - agents work at inhuman speed, so sprints aren't time-boxed. They're batches of work between CEO decision points. You decide when the work is good enough to ship.
+- **Consultants as perspectives** - hiring a consultant doesn't fill a capability gap (they're all Claude underneath). It adds a focused lens on a different slice of the problem. Diverse perspectives produce better work.
+
+### Mechanical, Not Magical
+
+Agents in ODD don't chat with each other. There are no multi-turn conversations between LLMs, no "Agent A said X so Agent B responds Y." That's theater.
+
+Instead, coordination is almost entirely mechanical. Agents communicate through artifacts:
+
+- **Git diffs** tell the next agent what changed
+- **Test results** (pass/fail, stdout, tracebacks) are the feedback loop
+- **Files on disk** are the shared state
+
+Each agent gets a system prompt, the relevant files, and a task. It produces output. That output is evaluated mechanically (tests pass or they don't, diffs apply or they don't). The next agent gets the *result*, not a narrative. This is why organizational patterns work here, they were already designed to coordinate people who can't read each other's minds.
+
+The one exception is adversarial code review. An anonymous reviewer scores the code at the end of a sprint, writes a review file, and your lead dev reads it and fixes issues, looping until the score clears the bar or you move on. Even here, agents don't converse; they pass artifacts through files.
+
+## What It Looks Like
+
+```
+ODD - What would you like to do?
+
+  Work
+     1. Start a sprint
+     2. Go on vacation
+     3. Check project status
+     4. Run a standup
+  People
+     5. Talk to a team member
+     6. Run a retrospective
+  Research
+     7. Review the backlog
+     8. Research a topic (spike)
+     9. Get user feedback (beta test)
+  Organization
+    10. Set company values
+    11. Set budget
+    12. Manage the team (hire, fire, review)
+
+    13. Call it a day
+
+CEO> _
+```
+
+You pick a number. You talk to your team in plain English. They write real code to real files. You watch it happen in the war room, a live grid showing every agent's status, output, and tool calls. Jump in with a CEO interrupt when you need to course-correct. Or just go on vacation and let them run.
+
+## The Team
+
+| Role | Name | What They Do |
+|------|------|-------------|
+| Lead Dev | Cody | Architecture, implementation, technical decisions |
+| QA | Vera | Writes tests first, validates everything |
+| Recruiter | Scout | Finds and hires specialist consultants |
+| Secretary | Paige | Standup notes, onboarding docs, retro synthesis |
+| CFO | Bill | Tracks token costs, enforces budgets (deterministic, not an LLM) |
+
+Senior roles (Lead Dev, QA) default to Claude Opus. Other agents default to Sonnet. Bill isn't an agent, he's pure accounting logic. You configure model tiers per-role to control quality vs. cost; it's a management decision.
+
+## Install
+
+```bash
+pip install odd-org
+```
+
+ODD calls the Anthropic API directly. Get an API key from the [Anthropic Console](https://console.anthropic.com/) and set it:
+
+```bash
+export ANTHROPIC_API_KEY=sk-ant-...
+```
+
+Then just:
+
+```bash
+odd
+```
+
+That's it. The menu walks you through everything, including first-time setup.
+
+## Related Research
+
+This is currently far from where I envision it and there is still a lot of testing to do, but I believe I'm on the right track. I've been building this with Opus for a little more than a week and stumbled across a paper yesterday where now I might as well build it in public. CMU researchers published [Effective Strategies for Asynchronous Software Engineering Agents](https://arxiv.org/abs/2603.21489) ([code](https://github.com/JiayiGeng/CAID)), demonstrating that multi-agent coordination with centralized delegation, strict workspace isolation, and test-gated integration outperforms single-agent scaling by 14-27% on software engineering benchmarks.
+
+ODD shares the same core architecture - a central coordinator delegating to isolated agents with TDD as the quality gate. The CAID findings reinforce the approach: agent isolation matters, delegation quality is the bottleneck, and throwing more iterations at a single agent hits a ceiling that structured teamwork breaks through. ODD extends this with specialized roles instead of identical workers, human-in-the-loop decision points at sprint boundaries, and persistent team identity across sessions.
+
+## Development
+
+```bash
+git clone https://github.com/ThePoetCoder/ODD.git
+cd odd
+pip install -e .
+pytest
+```
+
+For a full list of CLI subcommands, run `odd --help`.
+
+## License
+
+MIT

odd_org-0.1.0/README.md

@@ -0,0 +1,130 @@
+# ODD - Organization Driven Development; an Executable Metaphor
+
+**We live in odd times**
+
+## The Idea
+
+You don't write code. You don't write specs. You run a company.
+
+ODD organizes AI agents like a real software team:
+* Lead Dev
+* QA
+* Recruiter
+* Secretary
+* CFO
+* Specialist Consultants
+
+They all collaborate through TDD and agile processes to build software. You provide vision and make decisions at sprint boundaries. They do everything else.
+
+This is the next step up the abstraction ladder:
+
+1. **You write code** to solve your problem
+2. **You write specs** for an agent to code
+3. **You direct a company of agents** and they spec, build, test, and ship.
+
+## Why Organizational Patterns?
+
+Sprints, standups, TDD, code review, retrospectives - these aren't just corporate rituals. They're coordination algorithms refined over decades of real software teams solving real problems. They evolved to handle communication overhead, context switching, uncertainty, and quality control.
+
+AI agents have the same coordination problems. ODD applies the patterns that already work:
+
+- **Strict agent isolation** - each team member is a separate API call with their own persona. No shared memory, no context bleed between roles. The same reason you don't want your QA engineer thinking like your developer.
+- **TDD as quality gate** - QA writes tests *before* implementation. Tests passing is a more objective quality signal than code review alone, but you can add code review too.
+- **Scope-boxed sprints** - agents work at inhuman speed, so sprints aren't time-boxed. They're batches of work between CEO decision points. You decide when the work is good enough to ship.
+- **Consultants as perspectives** - hiring a consultant doesn't fill a capability gap (they're all Claude underneath). It adds a focused lens on a different slice of the problem. Diverse perspectives produce better work.
+
+### Mechanical, Not Magical
+
+Agents in ODD don't chat with each other. There are no multi-turn conversations between LLMs, no "Agent A said X so Agent B responds Y." That's theater.
+
+Instead, coordination is almost entirely mechanical. Agents communicate through artifacts:
+
+- **Git diffs** tell the next agent what changed
+- **Test results** (pass/fail, stdout, tracebacks) are the feedback loop
+- **Files on disk** are the shared state
+
+Each agent gets a system prompt, the relevant files, and a task. It produces output. That output is evaluated mechanically (tests pass or they don't, diffs apply or they don't). The next agent gets the *result*, not a narrative. This is why organizational patterns work here, they were already designed to coordinate people who can't read each other's minds.
+
+The one exception is adversarial code review. An anonymous reviewer scores the code at the end of a sprint, writes a review file, and your lead dev reads it and fixes issues, looping until the score clears the bar or you move on. Even here, agents don't converse; they pass artifacts through files.
+
+## What It Looks Like
+
+```
+ODD - What would you like to do?
+
+  Work
+     1. Start a sprint
+     2. Go on vacation
+     3. Check project status
+     4. Run a standup
+  People
+     5. Talk to a team member
+     6. Run a retrospective
+  Research
+     7. Review the backlog
+     8. Research a topic (spike)
+     9. Get user feedback (beta test)
+  Organization
+    10. Set company values
+    11. Set budget
+    12. Manage the team (hire, fire, review)
+
+    13. Call it a day
+
+CEO> _
+```
+
+You pick a number. You talk to your team in plain English. They write real code to real files. You watch it happen in the war room, a live grid showing every agent's status, output, and tool calls. Jump in with a CEO interrupt when you need to course-correct. Or just go on vacation and let them run.
+
+## The Team
+
+| Role | Name | What They Do |
+|------|------|-------------|
+| Lead Dev | Cody | Architecture, implementation, technical decisions |
+| QA | Vera | Writes tests first, validates everything |
+| Recruiter | Scout | Finds and hires specialist consultants |
+| Secretary | Paige | Standup notes, onboarding docs, retro synthesis |
+| CFO | Bill | Tracks token costs, enforces budgets (deterministic, not an LLM) |
+
+Senior roles (Lead Dev, QA) default to Claude Opus. Other agents default to Sonnet. Bill isn't an agent, he's pure accounting logic. You configure model tiers per-role to control quality vs. cost; it's a management decision.
+
+## Install
+
+```bash
+pip install odd-org
+```
+
+ODD calls the Anthropic API directly. Get an API key from the [Anthropic Console](https://console.anthropic.com/) and set it:
+
+```bash
+export ANTHROPIC_API_KEY=sk-ant-...
+```
+
+Then just:
+
+```bash
+odd
+```
+
+That's it. The menu walks you through everything, including first-time setup.
+
+## Related Research
+
+This is currently far from where I envision it and there is still a lot of testing to do, but I believe I'm on the right track. I've been building this with Opus for a little more than a week and stumbled across a paper yesterday where now I might as well build it in public. CMU researchers published [Effective Strategies for Asynchronous Software Engineering Agents](https://arxiv.org/abs/2603.21489) ([code](https://github.com/JiayiGeng/CAID)), demonstrating that multi-agent coordination with centralized delegation, strict workspace isolation, and test-gated integration outperforms single-agent scaling by 14-27% on software engineering benchmarks.
+
+ODD shares the same core architecture - a central coordinator delegating to isolated agents with TDD as the quality gate. The CAID findings reinforce the approach: agent isolation matters, delegation quality is the bottleneck, and throwing more iterations at a single agent hits a ceiling that structured teamwork breaks through. ODD extends this with specialized roles instead of identical workers, human-in-the-loop decision points at sprint boundaries, and persistent team identity across sessions.
+
+## Development
+
+```bash
+git clone https://github.com/ThePoetCoder/ODD.git
+cd odd
+pip install -e .
+pytest
+```
+
+For a full list of CLI subcommands, run `odd --help`.
+
+## License
+
+MIT

odd_org-0.1.0/pyproject.toml

@@ -0,0 +1,66 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "odd-org"
+version = "0.1.0"
+description = "Organization Driven Development - AI agents organized like a real software company"
+readme = "README.md"
+requires-python = ">=3.11"
+license = "MIT"
+keywords = ["ai", "agents", "cli", "development", "organization", "claude", "anthropic"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Environment :: Console",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Software Development",
+    "Topic :: Software Development :: Testing",
+]
+dependencies = [
+    "anthropic>=0.40.0",
+    "click>=8.1.0",
+    "rich>=13.0.0",
+    "pydantic>=2.0.0",
+]
+
+[project.optional-dependencies]
+openai-compat = ["requests>=2.28.0"]
+
+[project.urls]
+Repository = "https://github.com/ThePoetCoder/odd"
+Issues = "https://github.com/ThePoetCoder/odd/issues"
+
+[project.scripts]
+odd = "odd.cli:main"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/odd"]
+
+[tool.bandit]
+exclude_dirs = ["tests"]
+skips = ["B404"]  # import_subprocess is expected - we use subprocess intentionally
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+pythonpath = ["src"]
+markers = [
+    "integration: tests that hit the real Anthropic API (use -m integration to run)",
+]
+addopts = "-m 'not integration'"
+
+[tool.coverage.run]
+source = ["src/odd"]
+
+[tool.coverage.report]
+exclude_lines = [
+    "pragma: no cover",
+    "if __name__ == .__main__.",
+    "if TYPE_CHECKING:",
+]
+show_missing = true

odd_org-0.1.0/requirements.lock

@@ -0,0 +1,63 @@
+#
+# This file is autogenerated by pip-compile with Python 3.11
+# by the following command:
+#
+#    pip-compile --output-file=requirements.lock --strip-extras pyproject.toml
+#
+annotated-types==0.7.0
+    # via pydantic
+anthropic==0.86.0
+    # via odd (pyproject.toml)
+anyio==4.13.0
+    # via
+    #   anthropic
+    #   httpx
+certifi==2026.2.25
+    # via
+    #   httpcore
+    #   httpx
+click==8.3.1
+    # via odd (pyproject.toml)
+colorama==0.4.6
+    # via click
+distro==1.9.0
+    # via anthropic
+docstring-parser==0.17.0
+    # via anthropic
+h11==0.16.0
+    # via httpcore
+httpcore==1.0.9
+    # via httpx
+httpx==0.28.1
+    # via anthropic
+idna==3.11
+    # via
+    #   anyio
+    #   httpx
+jiter==0.13.0
+    # via anthropic
+markdown-it-py==4.0.0
+    # via rich
+mdurl==0.1.2
+    # via markdown-it-py
+pydantic==2.12.5
+    # via
+    #   anthropic
+    #   odd (pyproject.toml)
+pydantic-core==2.41.5
+    # via pydantic
+pygments==2.19.2
+    # via rich
+rich==14.3.3
+    # via odd (pyproject.toml)
+sniffio==1.3.1
+    # via anthropic
+typing-extensions==4.15.0
+    # via
+    #   anthropic
+    #   anyio
+    #   pydantic
+    #   pydantic-core
+    #   typing-inspection
+typing-inspection==0.4.2
+    # via pydantic

odd_org-0.1.0/src/odd/__init__.py

@@ -0,0 +1,3 @@
+"""ODD - Organization Driven Development."""
+
+__version__ = "0.1.0"

odd_org-0.1.0/src/odd/__main__.py

@@ -0,0 +1,20 @@
+"""Allow running as python -m odd."""
+
+import random
+import sys
+
+from rich.console import Console
+
+
+def _run() -> None:
+    from odd.cli import CLOSING_REMARKS, main
+
+    try:
+        main()
+    except KeyboardInterrupt:
+        console = Console()
+        console.print(f"\n[bold yellow]{random.choice(CLOSING_REMARKS)}[/bold yellow]")
+        sys.exit(0)
+
+
+_run()