adversarial-workflow 0.6.6__tar.gz → 0.9.0__tar.gz

{adversarial_workflow-0.6.6 → adversarial_workflow-0.9.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: adversarial-workflow
-Version: 0.6.6
+Version: 0.9.0
 Summary: Multi-stage AI evaluation system for task plans, code review, and test validation
 Author: Fredrik Matheson
 License: MIT
@@ -24,9 +24,11 @@ License-File: LICENSE
 Requires-Dist: pyyaml>=6.0
 Requires-Dist: python-dotenv>=0.19.0
 Requires-Dist: aider-chat>=0.86.0
+Requires-Dist: aiohttp>=3.8.0
 Provides-Extra: dev
 Requires-Dist: pytest>=7.0; extra == "dev"
 Requires-Dist: pytest-cov>=3.0; extra == "dev"
+Requires-Dist: pytest-asyncio>=0.21.0; extra == "dev"
 Requires-Dist: black>=22.0; extra == "dev"
 Requires-Dist: isort>=5.0; extra == "dev"
 Requires-Dist: flake8>=4.0; extra == "dev"
@@ -55,7 +57,7 @@ Evaluate proposals, sort out ideas, and prevent "phantom work" (AI claiming to i
 - 🎯 **Tool-agnostic**: Use with Claude Code, Cursor, Aider, manual coding, or any workflow
 - ✨ **Interactive onboarding**: Guided setup wizard gets you started in <5 minutes
 
-## What's New in v0.6.3
+## What's New in v0.9.0
 
 ### Upgrade
 
@@ -63,6 +65,67 @@ Evaluate proposals, sort out ideas, and prevent "phantom work" (AI claiming to i
 pip install --upgrade adversarial-workflow
 ```
 
+### v0.9.0 - Run Library Evaluators
+
+**Finally run your installed evaluators!** Use the new `--evaluator` flag:
+
+```bash
+# Install an evaluator from the library
+adversarial library install google/gemini-flash
+
+# Run it with --evaluator flag
+adversarial evaluate --evaluator gemini-flash task.md
+adversarial evaluate -e gemini-flash task.md  # short form
+
+# Works with model_requirement for portable evaluators
+# Automatically resolves to best available model
+```
+
+**Key Features:**
+- Run any installed evaluator by name
+- Supports evaluator aliases
+- Automatic model resolution via `model_requirement`
+- Falls back to legacy `model` field if resolution fails
+- Full backward compatibility - no flag uses existing behavior
+
+See [Evaluator Library](#evaluator-library) for full documentation.
+
+### v0.8.1 - BugBot Fixes
+
+- **CI/CD compatibility**: `--category --dry-run` no longer hangs in non-TTY environments
+- **Proper exit codes**: Dry-run returns 1 when all previews fail
+- **Config robustness**: Non-dict YAML configs no longer crash
+
+### v0.7.0 - Evaluator Library
+
+Browse, install, and update evaluators from the community [adversarial-evaluator-library](https://github.com/movito/adversarial-evaluator-library):
+
+```bash
+# Browse available evaluators
+adversarial library list
+
+# Filter by provider or category
+adversarial library list --provider google
+adversarial library list --category quick-check
+
+# Install evaluators
+adversarial library install google/gemini-flash openai/fast-check
+
+# Check for updates
+adversarial library check-updates
+
+# Update installed evaluators
+adversarial library update --all
+```
+
+**Key Features:**
+- Index caching with 1-hour TTL for faster lookups
+- Offline support with stale cache fallback
+- Provenance tracking via `_meta` block in installed files
+- Diff preview before applying updates
+
+See [Evaluator Library](#evaluator-library) for full documentation.
+
 ### v0.6.3 - Configurable Timeouts
 
 - **Per-evaluator timeout**: Add `timeout: 300` to evaluator YAML for slow models like Mistral Large
@@ -427,7 +490,8 @@ adversarial health                      # Comprehensive system health check
 adversarial agent onboard               # Set up agent coordination system
 
 # Workflow
-adversarial evaluate task.md            # Phase 1: Evaluate plan
+adversarial evaluate task.md            # Phase 1: Evaluate plan (uses config.yml)
+adversarial evaluate -e <name> task.md  # Phase 1: Evaluate with installed evaluator
 adversarial split task.md               # Split large files into smaller parts
 adversarial split task.md --dry-run     # Preview split without creating files
 adversarial review                      # Phase 3: Review implementation
@@ -435,6 +499,99 @@ adversarial validate "pytest"           # Phase 4: Validate with tests
 adversarial list-evaluators             # List all available evaluators
 ```
 
+## Evaluator Library
+
+Browse and install pre-configured evaluators from the community [adversarial-evaluator-library](https://github.com/movito/adversarial-evaluator-library).
+
+### Quick Start
+
+```bash
+# Browse available evaluators
+adversarial library list
+
+# Filter by provider or category
+adversarial library list --provider google
+adversarial library list --category quick-check
+
+# Install an evaluator
+adversarial library install google/gemini-flash
+
+# Run it with --evaluator flag
+adversarial evaluate --evaluator gemini-flash task.md
+adversarial evaluate -e gemini-flash task.md  # short form
+```
+
+### Available Commands
+
+| Command | Description |
+|---------|-------------|
+| `adversarial library list` | Browse available evaluators |
+| `adversarial library install <provider>/<name>` | Install evaluator to project |
+| `adversarial library check-updates` | Check for updates to installed evaluators |
+| `adversarial library update <name>` | Update an evaluator (with diff preview) |
+
+### Running Installed Evaluators
+
+Use the `--evaluator` flag to run any installed evaluator:
+
+```bash
+# Run by name
+adversarial evaluate --evaluator plan-evaluator task.md
+
+# Short form
+adversarial evaluate -e security-reviewer task.md
+
+# Evaluators with model_requirement auto-resolve to best available model
+adversarial evaluate -e gemini-flash task.md
+```
+
+**How it works:**
+- Looks up evaluator in `.adversarial/evaluators/*.yml`
+- Uses the evaluator's model, prompt, and output settings
+- Supports evaluator aliases
+- If evaluator has `model_requirement`, resolves to best available model
+- Falls back to legacy `model` field if resolution fails
+
+**Without --evaluator flag**: Uses existing shell script behavior (backward compatible)
+
+### Philosophy: Copy, Don't Link
+
+Installed evaluators are **copied** to your project, not referenced at runtime:
+- Projects remain self-contained and work offline
+- You can customize your local copies freely
+- Updates are explicit and user-controlled
+
+### Provenance Tracking
+
+Installed evaluators include metadata for tracking updates:
+
+```yaml
+_meta:
+  source: adversarial-evaluator-library
+  source_path: google/gemini-flash
+  version: "1.2.0"
+  installed: "2026-02-03T10:00:00Z"
+
+name: gemini-flash
+# ... rest of evaluator config
+```
+
+### Options
+
+```bash
+# Bypass cache (1-hour TTL by default)
+adversarial library list --no-cache
+
+# Force overwrite existing files
+adversarial library install google/gemini-flash --force
+
+# Update all outdated evaluators
+adversarial library update --all
+
+# Preview changes without applying
+adversarial library update gemini-flash --diff-only
+```
+
 ## Custom Evaluators
 
 Starting with v0.6.0, you can define project-specific evaluators without modifying the package.

{adversarial_workflow-0.6.6 → adversarial_workflow-0.9.0}/README.md

@@ -20,7 +20,7 @@ Evaluate proposals, sort out ideas, and prevent "phantom work" (AI claiming to i
 - 🎯 **Tool-agnostic**: Use with Claude Code, Cursor, Aider, manual coding, or any workflow
 - ✨ **Interactive onboarding**: Guided setup wizard gets you started in <5 minutes
 
-## What's New in v0.6.3
+## What's New in v0.9.0
 
 ### Upgrade
 
@@ -28,6 +28,67 @@ Evaluate proposals, sort out ideas, and prevent "phantom work" (AI claiming to i
 pip install --upgrade adversarial-workflow
 ```
 
+### v0.9.0 - Run Library Evaluators
+
+**Finally run your installed evaluators!** Use the new `--evaluator` flag:
+
+```bash
+# Install an evaluator from the library
+adversarial library install google/gemini-flash
+
+# Run it with --evaluator flag
+adversarial evaluate --evaluator gemini-flash task.md
+adversarial evaluate -e gemini-flash task.md  # short form
+
+# Works with model_requirement for portable evaluators
+# Automatically resolves to best available model
+```
+
+**Key Features:**
+- Run any installed evaluator by name
+- Supports evaluator aliases
+- Automatic model resolution via `model_requirement`
+- Falls back to legacy `model` field if resolution fails
+- Full backward compatibility - no flag uses existing behavior
+
+See [Evaluator Library](#evaluator-library) for full documentation.
+
+### v0.8.1 - BugBot Fixes
+
+- **CI/CD compatibility**: `--category --dry-run` no longer hangs in non-TTY environments
+- **Proper exit codes**: Dry-run returns 1 when all previews fail
+- **Config robustness**: Non-dict YAML configs no longer crash
+
+### v0.7.0 - Evaluator Library
+
+Browse, install, and update evaluators from the community [adversarial-evaluator-library](https://github.com/movito/adversarial-evaluator-library):
+
+```bash
+# Browse available evaluators
+adversarial library list
+
+# Filter by provider or category
+adversarial library list --provider google
+adversarial library list --category quick-check
+
+# Install evaluators
+adversarial library install google/gemini-flash openai/fast-check
+
+# Check for updates
+adversarial library check-updates
+
+# Update installed evaluators
+adversarial library update --all
+```
+
+**Key Features:**
+- Index caching with 1-hour TTL for faster lookups
+- Offline support with stale cache fallback
+- Provenance tracking via `_meta` block in installed files
+- Diff preview before applying updates
+
+See [Evaluator Library](#evaluator-library) for full documentation.
+
 ### v0.6.3 - Configurable Timeouts
 
 - **Per-evaluator timeout**: Add `timeout: 300` to evaluator YAML for slow models like Mistral Large
@@ -392,7 +453,8 @@ adversarial health                      # Comprehensive system health check
 adversarial agent onboard               # Set up agent coordination system
 
 # Workflow
-adversarial evaluate task.md            # Phase 1: Evaluate plan
+adversarial evaluate task.md            # Phase 1: Evaluate plan (uses config.yml)
+adversarial evaluate -e <name> task.md  # Phase 1: Evaluate with installed evaluator
 adversarial split task.md               # Split large files into smaller parts
 adversarial split task.md --dry-run     # Preview split without creating files
 adversarial review                      # Phase 3: Review implementation
@@ -400,6 +462,99 @@ adversarial validate "pytest"           # Phase 4: Validate with tests
 adversarial list-evaluators             # List all available evaluators
 ```
 
+## Evaluator Library
+
+Browse and install pre-configured evaluators from the community [adversarial-evaluator-library](https://github.com/movito/adversarial-evaluator-library).
+
+### Quick Start
+
+```bash
+# Browse available evaluators
+adversarial library list
+
+# Filter by provider or category
+adversarial library list --provider google
+adversarial library list --category quick-check
+
+# Install an evaluator
+adversarial library install google/gemini-flash
+
+# Run it with --evaluator flag
+adversarial evaluate --evaluator gemini-flash task.md
+adversarial evaluate -e gemini-flash task.md  # short form
+```
+
+### Available Commands
+
+| Command | Description |
+|---------|-------------|
+| `adversarial library list` | Browse available evaluators |
+| `adversarial library install <provider>/<name>` | Install evaluator to project |
+| `adversarial library check-updates` | Check for updates to installed evaluators |
+| `adversarial library update <name>` | Update an evaluator (with diff preview) |
+
+### Running Installed Evaluators
+
+Use the `--evaluator` flag to run any installed evaluator:
+
+```bash
+# Run by name
+adversarial evaluate --evaluator plan-evaluator task.md
+
+# Short form
+adversarial evaluate -e security-reviewer task.md
+
+# Evaluators with model_requirement auto-resolve to best available model
+adversarial evaluate -e gemini-flash task.md
+```
+
+**How it works:**
+- Looks up evaluator in `.adversarial/evaluators/*.yml`
+- Uses the evaluator's model, prompt, and output settings
+- Supports evaluator aliases
+- If evaluator has `model_requirement`, resolves to best available model
+- Falls back to legacy `model` field if resolution fails
+
+**Without --evaluator flag**: Uses existing shell script behavior (backward compatible)
+
+### Philosophy: Copy, Don't Link
+
+Installed evaluators are **copied** to your project, not referenced at runtime:
+- Projects remain self-contained and work offline
+- You can customize your local copies freely
+- Updates are explicit and user-controlled
+
+### Provenance Tracking
+
+Installed evaluators include metadata for tracking updates:
+
+```yaml
+_meta:
+  source: adversarial-evaluator-library
+  source_path: google/gemini-flash
+  version: "1.2.0"
+  installed: "2026-02-03T10:00:00Z"
+
+name: gemini-flash
+# ... rest of evaluator config
+```
+
+### Options
+
+```bash
+# Bypass cache (1-hour TTL by default)
+adversarial library list --no-cache
+
+# Force overwrite existing files
+adversarial library install google/gemini-flash --force
+
+# Update all outdated evaluators
+adversarial library update --all
+
+# Preview changes without applying
+adversarial library update gemini-flash --diff-only
+```
+
 ## Custom Evaluators
 
 Starting with v0.6.0, you can define project-specific evaluators without modifying the package.

{adversarial_workflow-0.6.6 → adversarial_workflow-0.9.0}/adversarial_workflow/__init__.py

@@ -12,7 +12,7 @@ Usage:
     adversarial validate "pytest"
 """
 
-__version__ = "0.6.6"
+__version__ = "0.9.0"
 __author__ = "Fredrik Matheson"
 __license__ = "MIT"