dissenter 1.0.2__tar.gz

dissenter-1.0.2/.github/workflows/publish.yml

@@ -0,0 +1,31 @@
+name: Publish to PyPI
+
+on:
+  push:
+    tags:
+      - "v*"
+  workflow_dispatch:
+
+jobs:
+  publish:
+    name: Build and publish
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      id-token: write  # required for trusted publishing (OIDC — no token needed)
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+
+      - name: Build
+        run: uv build
+
+      - name: Publish to PyPI
+        run: uv publish
+        # Uses OIDC trusted publishing — configure at:
+        # https://pypi.org/manage/project/dissenter/settings/publishing/
+        # Set publisher: GitHub, repo: PR0CK0/dissenter, workflow: publish.yml

dissenter-1.0.2/.gitignore

@@ -0,0 +1,11 @@
+decisions/
+.venv/
+__pycache__/
+*.pyc
+*.pyo
+.python-version
+dist/
+*.egg-info/
+.DS_Store
+.pytest_cache/
+.claude/

dissenter-1.0.2/CHANGELOG.md

@@ -0,0 +1,114 @@
+# Changelog
+
+All notable changes to this project are documented here.
+
+## [1.0.2] — 2026-03-21
+
+### Added
+- `make publish` target for manual PyPI deploys (`PYPI_TOKEN=pypi-xxxx make publish`).
+- GitHub Actions `workflow_dispatch` trigger — publish manually from the GitHub Actions UI without a tag push.
+
+---
+
+## [1.0.1] — 2026-03-21
+
+### Changed
+- Renamed project/package/CLI from `dissent` to `dissenter` (PyPI name `dissent` was taken).
+  - PyPI package: `pip install dissenter`
+  - CLI command: `dissenter ask / init / models / show`
+  - Default config file: `dissenter.toml` (was `dissent.toml`)
+  - Config dir: `~/.config/dissenter/` (was `~/.config/dissent/`)
+  - GitHub repo renamed to `PR0CK0/dissenter`
+  - Python import path unchanged (`from dissent.xxx import ...`)
+- Bumped version to 1.0.1.
+
+---
+
+## [1.0.0] — 2026-03-21
+
+### Added
+- **`dissent init`**: Interactive config wizard. Detects installed Ollama models and
+  claude/gemini CLI tools, prompts for rounds/models/roles/auth, previews generated
+  TOML, and saves to `dissent.toml`. Supports both chairman and dual-arbiter finals.
+- **`dissent models`**: Shows detected Ollama models, CLI tool paths, and API provider
+  key status (which env vars are set) in one command.
+- **`dissent ask --model` / `--chairman` / `--quick`**: Run debates without a config file.
+  `--model model_id[@role]` (repeatable) builds a debate round inline.
+  `--quick` auto-detects all installed Ollama models and runs immediately.
+- **PyPI publishing**: GitHub Actions workflow triggers on version tags and publishes
+  via `uv publish` with OIDC trusted publishing (no token management needed).
+- **`detect.py`**: Shared environment detection utilities (`detect_ollama_models`,
+  `detect_clis`, `detect_api_keys`, `infer_auth`).
+
+### Changed
+- `dissent show` now displays auth mode alongside each model.
+- Config priority for `ask`: `--quick` > `--model/--chairman` > config file.
+
+---
+
+## [0.2.1] — 2026-03-21
+
+### Changed
+- Output structure unified: all files for a run now live under
+  `decisions/<timestamp>/`. The decision is `decision.md` and per-round
+  debug files are subdirectories within the same folder. Previously the
+  decision file and the debug directory were siblings with different names.
+
+---
+
+## [0.2.0] — 2026-03-21
+
+### Added
+- **CLI auth mode** (`auth = "cli"`): Each model can now authenticate via a locally-installed
+  provider CLI instead of an API key. Auto-detects `claude` for Anthropic and `gemini` for
+  Google/Gemini providers. Override with `cli_command = "..."` in config.
+- **Dual-arbiter final round**: Final round can have 2 models (conservative + liberal roles)
+  combined side-by-side using a `combine_model`. Output is a `Dual Recommendation` document.
+- **Synthesis respects auth mode**: Synthesis phase now correctly routes through CLI or API
+  based on each model's `auth` field (was previously always using litellm).
+- **Error classification**: Friendly error messages for missing API keys, Ollama not running,
+  model not installed, rate limits, and context window exceeded.
+- **Clickable output links**: Terminal output uses OSC 8 hyperlinks to the saved decision file.
+- `dissent-test.toml`: Minimal test config using `ollama/ministral-3:3b` (no API keys needed).
+- `ensemble` binary removed from tracked files.
+
+### Changed
+- `dissent.toml`: Updated to use `auth = "cli"` for Anthropic and Gemini models.
+- `dissent.toml`: Fixed `qwen2.5:7b` → `qwen2.5:1.5b`.
+- `dissent.toml`: Fixed `ministral:3b` → `ministral-3:3b`.
+- CLI no longer prints full ADR to stdout; links to the saved file instead.
+
+### Fixed
+- Same model with different roles in one round: composite key `id::role::index` prevents
+  collision when the same model ID appears multiple times.
+- Synthesis `combine_model` (a string) is now wrapped in a `ModelConfig` for routing.
+
+---
+
+## [0.1.0] — 2026-03-18
+
+### Added
+- **Multi-round debate engine**: Arbitrary number of sequential rounds; models run in parallel
+  within each round and receive prior round context.
+- **Enforced final round**: Last `[[rounds]]` block must have exactly 1 model (chairman) or
+  2 models with a `combine_model` (dual-arbiter). Validated at config load.
+- **Role prompt system**: 10 adversarial roles extracted to individual TOML files under
+  `src/dissent/roles/`. Roles: analyst, chairman, conservative, contrarian, devil's advocate,
+  liberal, pragmatist, researcher, second opinion, skeptic.
+- **ADR output**: Chairman synthesis produces a structured Architectural Decision Record with
+  Context, Consensus, Disagreements, Options, Decision, Consequences, and Open Questions.
+- **Rich live status display**: Per-round tables showing model, role, elapsed time, and
+  word count / error for each model as they complete.
+- **Config validation**: pydantic v2 models with `model_validator` enforcing round constraints.
+- **`dissent ask`**: Run a debate and save output to `decisions/` with timestamped filenames.
+- **`dissent show`**: List saved decisions and open one interactively.
+- **Debug dir**: Per-run `debug/` subdirectory with raw round outputs for inspection.
+- **Unit and integration tests**: 23 tests across config, roles, runner, and integration.
+- **Makefile**: `ask`, `ask-test`, `show`, `install`, `test` targets.
+- **README**: Full documentation with architecture diagram, competitor comparison, install
+  guide, config reference, role catalog, and academic foundations.
+
+### Project
+- Renamed from `llm-ensemble` to `dissent`.
+- Package: `dissent` v0.1.0 → v0.2.0 (single release includes all above).
+- Requires Python 3.11+, uv, litellm, typer, rich, pydantic, platformdirs.

dissenter-1.0.2/Makefile

@@ -0,0 +1,22 @@
+.PHONY: ask ask-test show install test publish
+
+UV := $(shell command -v uv 2>/dev/null || echo $(HOME)/.local/bin/uv)
+
+ask:
+	@$(UV) run dissenter ask "$(Q)"
+
+ask-test:
+	@$(UV) run dissenter ask "$(Q)" --config dissenter-test.toml
+
+show:
+	@$(UV) run dissenter show
+
+install:
+	@$(UV) sync
+
+test:
+	@$(UV) run pytest tests/ -v
+
+publish:
+	@$(UV) build
+	@UV_PUBLISH_TOKEN=$(PYPI_TOKEN) $(UV) publish

dissenter-1.0.2/PKG-INFO

@@ -0,0 +1,13 @@
+Metadata-Version: 2.4
+Name: dissenter
+Version: 1.0.2
+Summary: Multi-LLM debate engine for architectural decisions
+Requires-Python: >=3.11
+Requires-Dist: litellm>=1.62.0
+Requires-Dist: platformdirs>=4.3.0
+Requires-Dist: pydantic>=2.9.0
+Requires-Dist: rich>=13.9.0
+Requires-Dist: typer>=0.12.0
+Provides-Extra: dev
+Requires-Dist: pytest-asyncio>=0.24; extra == 'dev'
+Requires-Dist: pytest>=8.0; extra == 'dev'