sugarai 3.1.1.dev6__tar.gz → 3.2.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/PKG-INFO +107 -1
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/README.md +102 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/pyproject.toml +6 -1
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/executor/agent_sdk_executor.py +130 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/main.py +91 -1
- sugarai-3.2.0/sugar/mcp/__init__.py +34 -0
- sugarai-3.2.0/sugar/ralph/__init__.py +19 -0
- sugarai-3.2.0/sugar/ralph/config.py +75 -0
- sugarai-3.2.0/sugar/ralph/profile.py +365 -0
- sugarai-3.2.0/sugar/ralph/validator.py +293 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugarai.egg-info/PKG-INFO +107 -1
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugarai.egg-info/SOURCES.txt +7 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugarai.egg-info/requires.txt +5 -0
- sugarai-3.2.0/tests/test_github_client.py +561 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_goose_integration.py +3 -2
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_issue_responder_cli.py +36 -7
- sugarai-3.2.0/tests/test_mcp_server.py +253 -0
- sugarai-3.2.0/tests/test_ralph.py +410 -0
- sugarai-3.1.1.dev6/sugar/mcp/__init__.py +0 -11
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/LICENSE +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/setup.cfg +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/__version__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/agent/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/agent/base.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/agent/hooks.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/agent/subagent_manager.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/agent/tools.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/billing/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/billing/api_keys.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/billing/tiers.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/billing/usage.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/config/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/config/issue_responder_config.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/config/prompt_config.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/core/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/core/loop.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/discovery/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/discovery/code_quality.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/discovery/error_monitor.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/discovery/github_watcher.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/discovery/test_coverage.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/executor/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/executor/base.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/executor/claude_wrapper.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/executor/structured_request.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/integrations/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/integrations/github.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/learning/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/learning/adaptive_scheduler.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/learning/feedback_processor.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/mcp/server.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/orchestration/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/orchestration/agent_router.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/orchestration/task_orchestrator.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/profiles/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/profiles/base.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/profiles/default.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/profiles/issue_responder.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/quality_gates/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/quality_gates/coordinator.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/quality_gates/diff_validator.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/quality_gates/evidence.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/quality_gates/failure_handler.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/quality_gates/functional_verifier.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/quality_gates/preflight_checks.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/quality_gates/success_criteria.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/quality_gates/test_validator.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/quality_gates/truth_enforcer.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/storage/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/storage/issue_response_manager.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/storage/task_type_manager.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/storage/work_queue.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/utils/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/utils/git_operations.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/utils/toon_encoder.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/workflow/__init__.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugar/workflow/orchestrator.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugarai.egg-info/dependency_links.txt +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugarai.egg-info/entry_points.txt +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/sugarai.egg-info/top_level.txt +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_agent.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_billing.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_cli.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_core_loop.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_hold_functionality.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_hooks.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_issue_response_integration.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_orchestration.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_profiles.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_prompt_config.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_quality_gates.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_storage.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_subagent_manager.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_task_types.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_toon_encoder.py +0 -0
- {sugarai-3.1.1.dev6 → sugarai-3.2.0}/tests/test_v3_benchmarks.py +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: sugarai
|
|
3
|
-
Version: 3.
|
|
3
|
+
Version: 3.2.0
|
|
4
4
|
Summary: Sugar 🍰 - A dev team that never stops. Delegate full tasks to AI in the background.
|
|
5
5
|
Author-email: Steven Leggett <contact@roboticforce.io>
|
|
6
6
|
License: MIT
|
|
@@ -50,6 +50,10 @@ Provides-Extra: test
|
|
|
50
50
|
Requires-Dist: pytest>=7.0.0; extra == "test"
|
|
51
51
|
Requires-Dist: pytest-asyncio>=0.21.0; extra == "test"
|
|
52
52
|
Requires-Dist: pytest-cov>=4.0.0; extra == "test"
|
|
53
|
+
Provides-Extra: mcp
|
|
54
|
+
Requires-Dist: mcp>=1.0.0; extra == "mcp"
|
|
55
|
+
Requires-Dist: starlette>=0.27.0; extra == "mcp"
|
|
56
|
+
Requires-Dist: uvicorn>=0.22.0; extra == "mcp"
|
|
53
57
|
Dynamic: license-file
|
|
54
58
|
|
|
55
59
|
# Sugar 🍰
|
|
@@ -208,6 +212,108 @@ sugar list --status completed
|
|
|
208
212
|
- Confidence-based auto-posting
|
|
209
213
|
- Searchable issue history
|
|
210
214
|
|
|
215
|
+
**Ralph Wiggum Integration** *(New)*
|
|
216
|
+
- Iterative execution for complex tasks
|
|
217
|
+
- Self-correcting loops until tests pass
|
|
218
|
+
- Prevents single-shot failures
|
|
219
|
+
- Automatic completion detection
|
|
220
|
+
|
|
221
|
+
## Ralph Wiggum: Why Sugar Gets It Right
|
|
222
|
+
|
|
223
|
+
Here's the thing about AI coding: **single-shot attempts often fail on complex tasks**.
|
|
224
|
+
|
|
225
|
+
Ask Claude to implement a feature in one go, and you might get something that's 80% right. But that 20% means broken tests, edge cases missed, or subtle bugs. You end up going back and forth, manually iterating until it works.
|
|
226
|
+
|
|
227
|
+
**Ralph Wiggum fixes this by design.**
|
|
228
|
+
|
|
229
|
+
Instead of trying to complete a task perfectly the first time, Sugar can feed the same prompt repeatedly. Each iteration:
|
|
230
|
+
|
|
231
|
+
1. Claude sees its previous work in the files
|
|
232
|
+
2. Runs tests and sees what's failing
|
|
233
|
+
3. Fixes issues and improves the implementation
|
|
234
|
+
4. Repeats until the task is actually complete
|
|
235
|
+
|
|
236
|
+
```bash
|
|
237
|
+
# Without Ralph (traditional single-shot):
|
|
238
|
+
sugar add "Implement rate limiting"
|
|
239
|
+
# Claude attempts once, maybe tests fail, task marked "done" anyway
|
|
240
|
+
|
|
241
|
+
# With Ralph (iterative):
|
|
242
|
+
sugar add "Implement rate limiting" --ralph --max-iterations 10
|
|
243
|
+
# Claude iterates: implement → test → fix → test → fix → done
|
|
244
|
+
# Only marked complete when tests actually pass
|
|
245
|
+
```
|
|
246
|
+
|
|
247
|
+
**Think of it like code review cycles**, but automated. Junior dev writes code, tests fail, they fix it, tests pass, PR merged. Ralph does this loop automatically.
|
|
248
|
+
|
|
249
|
+
### When to Use Ralph
|
|
250
|
+
|
|
251
|
+
| Task Type | Without Ralph | With Ralph |
|
|
252
|
+
|-----------|---------------|------------|
|
|
253
|
+
| Simple bug fix | Works fine | Overkill |
|
|
254
|
+
| New feature | Hit or miss | Iterates until working |
|
|
255
|
+
| Complex refactor | Often breaks things | Self-corrects |
|
|
256
|
+
| TDD implementation | Tests often fail | Keeps going until green |
|
|
257
|
+
| Flaky test debugging | Might give up | Tries multiple approaches |
|
|
258
|
+
|
|
259
|
+
### How It Works
|
|
260
|
+
|
|
261
|
+
```
|
|
262
|
+
Iteration 1: "Implement rate limiting"
|
|
263
|
+
→ Creates RateLimiter class
|
|
264
|
+
→ Tests: 2 passing, 3 failing
|
|
265
|
+
|
|
266
|
+
Iteration 2: Same prompt, sees previous work
|
|
267
|
+
→ Fixes failing tests
|
|
268
|
+
→ Tests: 4 passing, 1 failing
|
|
269
|
+
|
|
270
|
+
Iteration 3: Same prompt, sees more progress
|
|
271
|
+
→ Handles edge case
|
|
272
|
+
→ Tests: 5 passing, 0 failing
|
|
273
|
+
→ Outputs: <promise>DONE</promise>
|
|
274
|
+
→ Task complete!
|
|
275
|
+
```
|
|
276
|
+
|
|
277
|
+
The `<promise>` tag is how Claude signals "I'm actually done." Without it, Ralph knows to keep iterating.
|
|
278
|
+
|
|
279
|
+
### Setup
|
|
280
|
+
|
|
281
|
+
**Ralph is built into Sugar** - no separate installation required. Just enable it in your config:
|
|
282
|
+
|
|
283
|
+
```yaml
|
|
284
|
+
# .sugar/config.yaml
|
|
285
|
+
sugar:
|
|
286
|
+
ralph:
|
|
287
|
+
enabled: true
|
|
288
|
+
max_iterations: 10
|
|
289
|
+
require_completion_criteria: true
|
|
290
|
+
```
|
|
291
|
+
|
|
292
|
+
Or use the `--ralph` flag when adding tasks:
|
|
293
|
+
|
|
294
|
+
```bash
|
|
295
|
+
sugar add "Complex feature" --ralph --max-iterations 15
|
|
296
|
+
```
|
|
297
|
+
|
|
298
|
+
### Safety First
|
|
299
|
+
|
|
300
|
+
Ralph won't run forever. You must include:
|
|
301
|
+
- A `<promise>` tag in your prompt (completion signal)
|
|
302
|
+
- OR `--max-iterations` flag (safety limit)
|
|
303
|
+
|
|
304
|
+
Sugar validates this BEFORE starting. No completion criteria = task rejected.
|
|
305
|
+
|
|
306
|
+
### Interactive Use (Claude Code)
|
|
307
|
+
|
|
308
|
+
For interactive Ralph loops in Claude Code sessions (outside of Sugar), install the Ralph Wiggum plugin:
|
|
309
|
+
|
|
310
|
+
```bash
|
|
311
|
+
# If you have the toolkit installed:
|
|
312
|
+
/ralph-wiggum:ralph-loop "Fix the flaky tests" --max-iterations 10
|
|
313
|
+
```
|
|
314
|
+
|
|
315
|
+
**Full docs:** [docs/ralph-wiggum.md](docs/ralph-wiggum.md)
|
|
316
|
+
|
|
211
317
|
## Issue Responder
|
|
212
318
|
|
|
213
319
|
Automatically analyze and respond to GitHub issues with AI-powered insights. Sugar understands issue context, codebase structure, and project patterns to generate helpful responses.
|
|
@@ -154,6 +154,108 @@ sugar list --status completed
|
|
|
154
154
|
- Confidence-based auto-posting
|
|
155
155
|
- Searchable issue history
|
|
156
156
|
|
|
157
|
+
**Ralph Wiggum Integration** *(New)*
|
|
158
|
+
- Iterative execution for complex tasks
|
|
159
|
+
- Self-correcting loops until tests pass
|
|
160
|
+
- Prevents single-shot failures
|
|
161
|
+
- Automatic completion detection
|
|
162
|
+
|
|
163
|
+
## Ralph Wiggum: Why Sugar Gets It Right
|
|
164
|
+
|
|
165
|
+
Here's the thing about AI coding: **single-shot attempts often fail on complex tasks**.
|
|
166
|
+
|
|
167
|
+
Ask Claude to implement a feature in one go, and you might get something that's 80% right. But that 20% means broken tests, edge cases missed, or subtle bugs. You end up going back and forth, manually iterating until it works.
|
|
168
|
+
|
|
169
|
+
**Ralph Wiggum fixes this by design.**
|
|
170
|
+
|
|
171
|
+
Instead of trying to complete a task perfectly the first time, Sugar can feed the same prompt repeatedly. Each iteration:
|
|
172
|
+
|
|
173
|
+
1. Claude sees its previous work in the files
|
|
174
|
+
2. Runs tests and sees what's failing
|
|
175
|
+
3. Fixes issues and improves the implementation
|
|
176
|
+
4. Repeats until the task is actually complete
|
|
177
|
+
|
|
178
|
+
```bash
|
|
179
|
+
# Without Ralph (traditional single-shot):
|
|
180
|
+
sugar add "Implement rate limiting"
|
|
181
|
+
# Claude attempts once, maybe tests fail, task marked "done" anyway
|
|
182
|
+
|
|
183
|
+
# With Ralph (iterative):
|
|
184
|
+
sugar add "Implement rate limiting" --ralph --max-iterations 10
|
|
185
|
+
# Claude iterates: implement → test → fix → test → fix → done
|
|
186
|
+
# Only marked complete when tests actually pass
|
|
187
|
+
```
|
|
188
|
+
|
|
189
|
+
**Think of it like code review cycles**, but automated. Junior dev writes code, tests fail, they fix it, tests pass, PR merged. Ralph does this loop automatically.
|
|
190
|
+
|
|
191
|
+
### When to Use Ralph
|
|
192
|
+
|
|
193
|
+
| Task Type | Without Ralph | With Ralph |
|
|
194
|
+
|-----------|---------------|------------|
|
|
195
|
+
| Simple bug fix | Works fine | Overkill |
|
|
196
|
+
| New feature | Hit or miss | Iterates until working |
|
|
197
|
+
| Complex refactor | Often breaks things | Self-corrects |
|
|
198
|
+
| TDD implementation | Tests often fail | Keeps going until green |
|
|
199
|
+
| Flaky test debugging | Might give up | Tries multiple approaches |
|
|
200
|
+
|
|
201
|
+
### How It Works
|
|
202
|
+
|
|
203
|
+
```
|
|
204
|
+
Iteration 1: "Implement rate limiting"
|
|
205
|
+
→ Creates RateLimiter class
|
|
206
|
+
→ Tests: 2 passing, 3 failing
|
|
207
|
+
|
|
208
|
+
Iteration 2: Same prompt, sees previous work
|
|
209
|
+
→ Fixes failing tests
|
|
210
|
+
→ Tests: 4 passing, 1 failing
|
|
211
|
+
|
|
212
|
+
Iteration 3: Same prompt, sees more progress
|
|
213
|
+
→ Handles edge case
|
|
214
|
+
→ Tests: 5 passing, 0 failing
|
|
215
|
+
→ Outputs: <promise>DONE</promise>
|
|
216
|
+
→ Task complete!
|
|
217
|
+
```
|
|
218
|
+
|
|
219
|
+
The `<promise>` tag is how Claude signals "I'm actually done." Without it, Ralph knows to keep iterating.
|
|
220
|
+
|
|
221
|
+
### Setup
|
|
222
|
+
|
|
223
|
+
**Ralph is built into Sugar** - no separate installation required. Just enable it in your config:
|
|
224
|
+
|
|
225
|
+
```yaml
|
|
226
|
+
# .sugar/config.yaml
|
|
227
|
+
sugar:
|
|
228
|
+
ralph:
|
|
229
|
+
enabled: true
|
|
230
|
+
max_iterations: 10
|
|
231
|
+
require_completion_criteria: true
|
|
232
|
+
```
|
|
233
|
+
|
|
234
|
+
Or use the `--ralph` flag when adding tasks:
|
|
235
|
+
|
|
236
|
+
```bash
|
|
237
|
+
sugar add "Complex feature" --ralph --max-iterations 15
|
|
238
|
+
```
|
|
239
|
+
|
|
240
|
+
### Safety First
|
|
241
|
+
|
|
242
|
+
Ralph won't run forever. You must include:
|
|
243
|
+
- A `<promise>` tag in your prompt (completion signal)
|
|
244
|
+
- OR `--max-iterations` flag (safety limit)
|
|
245
|
+
|
|
246
|
+
Sugar validates this BEFORE starting. No completion criteria = task rejected.
|
|
247
|
+
|
|
248
|
+
### Interactive Use (Claude Code)
|
|
249
|
+
|
|
250
|
+
For interactive Ralph loops in Claude Code sessions (outside of Sugar), install the Ralph Wiggum plugin:
|
|
251
|
+
|
|
252
|
+
```bash
|
|
253
|
+
# If you have the toolkit installed:
|
|
254
|
+
/ralph-wiggum:ralph-loop "Fix the flaky tests" --max-iterations 10
|
|
255
|
+
```
|
|
256
|
+
|
|
257
|
+
**Full docs:** [docs/ralph-wiggum.md](docs/ralph-wiggum.md)
|
|
258
|
+
|
|
157
259
|
## Issue Responder
|
|
158
260
|
|
|
159
261
|
Automatically analyze and respond to GitHub issues with AI-powered insights. Sugar understands issue context, codebase structure, and project patterns to generate helpful responses.
|
|
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "sugarai"
|
|
7
|
-
version = "3.
|
|
7
|
+
version = "3.2.0"
|
|
8
8
|
description = "Sugar 🍰 - A dev team that never stops. Delegate full tasks to AI in the background."
|
|
9
9
|
readme = "README.md"
|
|
10
10
|
requires-python = ">=3.11"
|
|
@@ -59,6 +59,11 @@ test = [
|
|
|
59
59
|
"pytest-asyncio>=0.21.0",
|
|
60
60
|
"pytest-cov>=4.0.0"
|
|
61
61
|
]
|
|
62
|
+
mcp = [
|
|
63
|
+
"mcp>=1.0.0",
|
|
64
|
+
"starlette>=0.27.0",
|
|
65
|
+
"uvicorn>=0.22.0",
|
|
66
|
+
]
|
|
62
67
|
|
|
63
68
|
[project.scripts]
|
|
64
69
|
sugar = "sugar.main:cli"
|
|
@@ -20,6 +20,7 @@ from ..storage import IssueResponseManager
|
|
|
20
20
|
from ..profiles import IssueResponderProfile
|
|
21
21
|
from ..config import IssueResponderConfig
|
|
22
22
|
from ..integrations import GitHubClient
|
|
23
|
+
from ..ralph import RalphWiggumProfile, RalphConfig
|
|
23
24
|
|
|
24
25
|
logger = logging.getLogger(__name__)
|
|
25
26
|
|
|
@@ -157,6 +158,129 @@ class AgentSDKExecutor(BaseExecutor):
|
|
|
157
158
|
finally:
|
|
158
159
|
await agent.end_session()
|
|
159
160
|
|
|
161
|
+
async def _execute_ralph_iteration(
|
|
162
|
+
self, work_item: Dict[str, Any]
|
|
163
|
+
) -> Dict[str, Any]:
|
|
164
|
+
"""Execute a work item with Ralph Wiggum iterative loop."""
|
|
165
|
+
context = work_item.get("context", {})
|
|
166
|
+
max_iterations = context.get("max_iterations", 10)
|
|
167
|
+
|
|
168
|
+
# Create Ralph config
|
|
169
|
+
ralph_config = RalphConfig(
|
|
170
|
+
max_iterations=max_iterations,
|
|
171
|
+
completion_promise=context.get("completion_promise", "DONE"),
|
|
172
|
+
require_completion_criteria=False, # Already validated at CLI
|
|
173
|
+
quality_gates_enabled=self.quality_gates_enabled,
|
|
174
|
+
stop_on_gate_failure=context.get("stop_on_gate_failure", False),
|
|
175
|
+
)
|
|
176
|
+
|
|
177
|
+
profile = RalphWiggumProfile(ralph_config=ralph_config)
|
|
178
|
+
|
|
179
|
+
# Validate input
|
|
180
|
+
input_data = {
|
|
181
|
+
"prompt": work_item.get("description", work_item.get("title", "")),
|
|
182
|
+
"work_item": work_item,
|
|
183
|
+
}
|
|
184
|
+
processed = await profile.process_input(input_data)
|
|
185
|
+
|
|
186
|
+
if not processed.get("valid", True):
|
|
187
|
+
return {
|
|
188
|
+
"success": False,
|
|
189
|
+
"error": "Invalid Ralph input: "
|
|
190
|
+
+ ", ".join(processed.get("errors", [])),
|
|
191
|
+
"timestamp": datetime.now(timezone.utc).isoformat(),
|
|
192
|
+
"work_item_id": work_item.get("id"),
|
|
193
|
+
}
|
|
194
|
+
|
|
195
|
+
start_time = datetime.now(timezone.utc)
|
|
196
|
+
agent = await self._get_agent()
|
|
197
|
+
prompt = processed.get("prompt", work_item.get("description", ""))
|
|
198
|
+
|
|
199
|
+
iteration_results = []
|
|
200
|
+
final_result = None
|
|
201
|
+
|
|
202
|
+
logger.info(f"Starting Ralph execution with max {max_iterations} iterations")
|
|
203
|
+
|
|
204
|
+
while profile.should_continue():
|
|
205
|
+
iteration = profile.current_iteration + 1 # 1-indexed for display
|
|
206
|
+
logger.info(f"Ralph iteration {iteration}/{max_iterations}")
|
|
207
|
+
|
|
208
|
+
try:
|
|
209
|
+
# Execute the prompt
|
|
210
|
+
response = await agent.execute(prompt)
|
|
211
|
+
|
|
212
|
+
# Process output to check for completion
|
|
213
|
+
output_data = {
|
|
214
|
+
"content": response.content,
|
|
215
|
+
"success": response.success,
|
|
216
|
+
"files_changed": getattr(response, "files_changed", []),
|
|
217
|
+
}
|
|
218
|
+
|
|
219
|
+
result = await profile.process_output(output_data)
|
|
220
|
+
|
|
221
|
+
iteration_results.append(
|
|
222
|
+
{
|
|
223
|
+
"iteration": iteration,
|
|
224
|
+
"complete": result.get("complete", False),
|
|
225
|
+
"promise_detected": result.get("promise_detected"),
|
|
226
|
+
"stuck": result.get("stuck", False),
|
|
227
|
+
}
|
|
228
|
+
)
|
|
229
|
+
|
|
230
|
+
if result.get("complete"):
|
|
231
|
+
logger.info(f"Ralph completed after {iteration} iterations")
|
|
232
|
+
final_result = {
|
|
233
|
+
"success": True,
|
|
234
|
+
"content": response.content,
|
|
235
|
+
"iterations": iteration,
|
|
236
|
+
"completion_reason": result.get(
|
|
237
|
+
"completion_reason", "promise_detected"
|
|
238
|
+
),
|
|
239
|
+
}
|
|
240
|
+
break
|
|
241
|
+
|
|
242
|
+
if result.get("stuck"):
|
|
243
|
+
logger.warning(
|
|
244
|
+
f"Ralph detected stuck state at iteration {iteration}"
|
|
245
|
+
)
|
|
246
|
+
final_result = {
|
|
247
|
+
"success": False,
|
|
248
|
+
"error": "Task appears stuck",
|
|
249
|
+
"iterations": iteration,
|
|
250
|
+
"stuck_reason": result.get("stuck_reason"),
|
|
251
|
+
}
|
|
252
|
+
break
|
|
253
|
+
|
|
254
|
+
except Exception as e:
|
|
255
|
+
logger.error(f"Ralph iteration {iteration} failed: {e}")
|
|
256
|
+
iteration_results.append(
|
|
257
|
+
{
|
|
258
|
+
"iteration": iteration,
|
|
259
|
+
"error": str(e),
|
|
260
|
+
}
|
|
261
|
+
)
|
|
262
|
+
|
|
263
|
+
# If loop ended without completion
|
|
264
|
+
if final_result is None:
|
|
265
|
+
final_result = {
|
|
266
|
+
"success": False,
|
|
267
|
+
"error": f"Max iterations ({max_iterations}) reached without completion",
|
|
268
|
+
"iterations": profile.current_iteration,
|
|
269
|
+
}
|
|
270
|
+
|
|
271
|
+
execution_time = (datetime.now(timezone.utc) - start_time).total_seconds()
|
|
272
|
+
stats = profile.get_iteration_stats()
|
|
273
|
+
|
|
274
|
+
return {
|
|
275
|
+
**final_result,
|
|
276
|
+
"timestamp": datetime.now(timezone.utc).isoformat(),
|
|
277
|
+
"work_item_id": work_item.get("id"),
|
|
278
|
+
"execution_time": execution_time,
|
|
279
|
+
"executor": "agent_sdk_ralph",
|
|
280
|
+
"ralph_stats": stats,
|
|
281
|
+
"iteration_results": iteration_results,
|
|
282
|
+
}
|
|
283
|
+
|
|
160
284
|
async def execute_work(self, work_item: Dict[str, Any]) -> Dict[str, Any]:
|
|
161
285
|
"""
|
|
162
286
|
Execute a work item using the Claude Agent SDK.
|
|
@@ -169,10 +293,16 @@ class AgentSDKExecutor(BaseExecutor):
|
|
|
169
293
|
"""
|
|
170
294
|
# Check for specialized task types
|
|
171
295
|
task_type = work_item.get("type", "")
|
|
296
|
+
context = work_item.get("context", {})
|
|
172
297
|
|
|
173
298
|
if task_type == "issue_response":
|
|
174
299
|
return await self._execute_issue_response(work_item)
|
|
175
300
|
|
|
301
|
+
# Check for Ralph-enabled tasks
|
|
302
|
+
if context.get("ralph_enabled"):
|
|
303
|
+
logger.info(f"Executing with Ralph Wiggum: {work_item.get('title')}")
|
|
304
|
+
return await self._execute_ralph_iteration(work_item)
|
|
305
|
+
|
|
176
306
|
if self.dry_run:
|
|
177
307
|
logger.info(f"DRY RUN: Simulating execution of {work_item.get('title')}")
|
|
178
308
|
return await self._simulate_execution(work_item)
|
|
@@ -311,6 +311,13 @@ def init(project_dir):
|
|
|
311
311
|
@click.option(
|
|
312
312
|
"--skip-stages", help="Comma-separated stages to skip (research,planning)"
|
|
313
313
|
)
|
|
314
|
+
@click.option("--ralph", is_flag=True, help="Enable Ralph Wiggum iterative execution")
|
|
315
|
+
@click.option(
|
|
316
|
+
"--max-iterations",
|
|
317
|
+
type=int,
|
|
318
|
+
default=None,
|
|
319
|
+
help="Max iterations for Ralph mode (default: 10)",
|
|
320
|
+
)
|
|
314
321
|
@click.pass_context
|
|
315
322
|
def add(
|
|
316
323
|
ctx,
|
|
@@ -325,6 +332,8 @@ def add(
|
|
|
325
332
|
parse_json,
|
|
326
333
|
orchestrate,
|
|
327
334
|
skip_stages,
|
|
335
|
+
ralph,
|
|
336
|
+
max_iterations,
|
|
328
337
|
):
|
|
329
338
|
"""Add a new task to Sugar work queue
|
|
330
339
|
|
|
@@ -415,6 +424,23 @@ def add(
|
|
|
415
424
|
s.strip() for s in skip_stages.split(",")
|
|
416
425
|
]
|
|
417
426
|
|
|
427
|
+
# Add Ralph Wiggum settings if enabled
|
|
428
|
+
if ralph:
|
|
429
|
+
task_data["context"]["ralph_enabled"] = True
|
|
430
|
+
task_data["context"]["max_iterations"] = max_iterations or 10
|
|
431
|
+
# Validate completion criteria if strict mode
|
|
432
|
+
from .ralph import CompletionCriteriaValidator
|
|
433
|
+
|
|
434
|
+
validator = CompletionCriteriaValidator(strict=False)
|
|
435
|
+
result = validator.validate(
|
|
436
|
+
description, {"max_iterations": max_iterations or 10}
|
|
437
|
+
)
|
|
438
|
+
if not result.is_valid:
|
|
439
|
+
click.echo(
|
|
440
|
+
f"⚠️ Ralph enabled but completion criteria may be weak: {', '.join(result.suggestions[:2])}",
|
|
441
|
+
err=True,
|
|
442
|
+
)
|
|
443
|
+
|
|
418
444
|
# Override/merge with complex input data
|
|
419
445
|
if task_data_override:
|
|
420
446
|
# Handle context merging specially to preserve base context
|
|
@@ -445,8 +471,13 @@ def add(
|
|
|
445
471
|
elif parse_json:
|
|
446
472
|
input_method = " (JSON parsed)"
|
|
447
473
|
|
|
474
|
+
ralph_mode = ""
|
|
475
|
+
if ralph:
|
|
476
|
+
max_iter = task_data["context"].get("max_iterations", 10)
|
|
477
|
+
ralph_mode = f" [Ralph: max {max_iter} iterations]"
|
|
478
|
+
|
|
448
479
|
click.echo(
|
|
449
|
-
f"✅ Added {task_data.get('type', task_type)} task: '{task_data.get('title', title)}' ({urgency}){input_method}"
|
|
480
|
+
f"✅ Added {task_data.get('type', task_type)} task: '{task_data.get('title', title)}' ({urgency}){input_method}{ralph_mode}"
|
|
450
481
|
)
|
|
451
482
|
|
|
452
483
|
except Exception as e:
|
|
@@ -2149,6 +2180,16 @@ sugar:
|
|
|
2149
2180
|
|
|
2150
2181
|
# Model selection
|
|
2151
2182
|
model: "claude-sonnet-4-20250514"
|
|
2183
|
+
|
|
2184
|
+
# Ralph Wiggum (iterative AI execution)
|
|
2185
|
+
ralph:
|
|
2186
|
+
enabled: true # Allow Ralph mode for tasks
|
|
2187
|
+
max_iterations: 10 # Default max iterations per task
|
|
2188
|
+
iteration_timeout: 300 # Max seconds per iteration (5 min)
|
|
2189
|
+
completion_promise: "DONE" # Default completion signal
|
|
2190
|
+
require_completion_criteria: true # Require <promise> tags or --max-iterations
|
|
2191
|
+
quality_gates_enabled: true # Run quality gates between iterations
|
|
2192
|
+
stop_on_gate_failure: false # Keep trying even if gates fail
|
|
2152
2193
|
"""
|
|
2153
2194
|
|
|
2154
2195
|
|
|
@@ -3386,5 +3427,54 @@ def issue_search(ctx, query, repo, limit):
|
|
|
3386
3427
|
sys.exit(1)
|
|
3387
3428
|
|
|
3388
3429
|
|
|
3430
|
+
# =============================================================================
|
|
3431
|
+
# MCP Server Commands
|
|
3432
|
+
# =============================================================================
|
|
3433
|
+
|
|
3434
|
+
|
|
3435
|
+
@cli.group()
|
|
3436
|
+
@click.pass_context
|
|
3437
|
+
def mcp(ctx):
|
|
3438
|
+
"""MCP (Model Context Protocol) server for Goose and other MCP clients"""
|
|
3439
|
+
pass
|
|
3440
|
+
|
|
3441
|
+
|
|
3442
|
+
@mcp.command("serve")
|
|
3443
|
+
@click.option("--host", default="0.0.0.0", help="Host to bind to")
|
|
3444
|
+
@click.option("--port", default=8080, type=int, help="Port to listen on")
|
|
3445
|
+
@click.option("--repo", help="Default repository (owner/repo)")
|
|
3446
|
+
@click.pass_context
|
|
3447
|
+
def mcp_serve(ctx, host, port, repo):
|
|
3448
|
+
"""Start the Sugar MCP server for Goose integration"""
|
|
3449
|
+
import asyncio
|
|
3450
|
+
|
|
3451
|
+
try:
|
|
3452
|
+
from .mcp.server import create_server
|
|
3453
|
+
except ImportError as e:
|
|
3454
|
+
click.echo(
|
|
3455
|
+
"❌ MCP dependencies not installed. Install with:\n"
|
|
3456
|
+
" pip install sugarai[mcp]",
|
|
3457
|
+
err=True,
|
|
3458
|
+
)
|
|
3459
|
+
click.echo(f"\nMissing: {e}", err=True)
|
|
3460
|
+
sys.exit(1)
|
|
3461
|
+
|
|
3462
|
+
click.echo(f"🍰 Starting Sugar MCP server on {host}:{port}")
|
|
3463
|
+
if repo:
|
|
3464
|
+
click.echo(f" Default repository: {repo}")
|
|
3465
|
+
click.echo("\nPress Ctrl+C to stop\n")
|
|
3466
|
+
|
|
3467
|
+
server = create_server(
|
|
3468
|
+
host=host,
|
|
3469
|
+
port=port,
|
|
3470
|
+
default_repo=repo,
|
|
3471
|
+
)
|
|
3472
|
+
|
|
3473
|
+
try:
|
|
3474
|
+
asyncio.run(server.run())
|
|
3475
|
+
except KeyboardInterrupt:
|
|
3476
|
+
click.echo("\n👋 Server stopped")
|
|
3477
|
+
|
|
3478
|
+
|
|
3389
3479
|
if __name__ == "__main__":
|
|
3390
3480
|
cli()
|
|
@@ -0,0 +1,34 @@
|
|
|
1
|
+
"""
|
|
2
|
+
Sugar MCP Server
|
|
3
|
+
|
|
4
|
+
Model Context Protocol server for Sugar, enabling integration with:
|
|
5
|
+
- GitHub Copilot Custom Agents
|
|
6
|
+
- Other MCP-compatible clients
|
|
7
|
+
|
|
8
|
+
Note: This module uses lazy imports to avoid requiring MCP dependencies
|
|
9
|
+
unless the MCP server is actually used. This allows the base Sugar CLI
|
|
10
|
+
to work without installing the optional [mcp] extras.
|
|
11
|
+
"""
|
|
12
|
+
|
|
13
|
+
__all__ = ["SugarMCPServer", "create_server"]
|
|
14
|
+
|
|
15
|
+
# Lazy import cache
|
|
16
|
+
_lazy_imports = {}
|
|
17
|
+
|
|
18
|
+
|
|
19
|
+
def __getattr__(name: str):
|
|
20
|
+
"""Lazy import to avoid requiring MCP dependencies unless actually used."""
|
|
21
|
+
if name in __all__:
|
|
22
|
+
if name not in _lazy_imports:
|
|
23
|
+
try:
|
|
24
|
+
from .server import SugarMCPServer, create_server
|
|
25
|
+
|
|
26
|
+
_lazy_imports["SugarMCPServer"] = SugarMCPServer
|
|
27
|
+
_lazy_imports["create_server"] = create_server
|
|
28
|
+
except ImportError as e:
|
|
29
|
+
raise ImportError(
|
|
30
|
+
f"MCP dependencies not installed. Install with: pip install sugarai[mcp]\n"
|
|
31
|
+
f"Original error: {e}"
|
|
32
|
+
) from e
|
|
33
|
+
return _lazy_imports[name]
|
|
34
|
+
raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
|
|
@@ -0,0 +1,19 @@
|
|
|
1
|
+
"""
|
|
2
|
+
Ralph Wiggum Integration - Iterative AI loop support for Sugar
|
|
3
|
+
|
|
4
|
+
This module provides:
|
|
5
|
+
- CompletionCriteriaValidator: Validates tasks have clear exit conditions
|
|
6
|
+
- RalphWiggumProfile: Profile for iterative task execution
|
|
7
|
+
- RalphConfig: Configuration for Ralph Wiggum loops
|
|
8
|
+
"""
|
|
9
|
+
|
|
10
|
+
from .validator import CompletionCriteriaValidator, ValidationResult
|
|
11
|
+
from .profile import RalphWiggumProfile
|
|
12
|
+
from .config import RalphConfig
|
|
13
|
+
|
|
14
|
+
__all__ = [
|
|
15
|
+
"CompletionCriteriaValidator",
|
|
16
|
+
"ValidationResult",
|
|
17
|
+
"RalphWiggumProfile",
|
|
18
|
+
"RalphConfig",
|
|
19
|
+
]
|
|
@@ -0,0 +1,75 @@
|
|
|
1
|
+
"""
|
|
2
|
+
Ralph Wiggum Configuration
|
|
3
|
+
|
|
4
|
+
Configuration for Ralph Wiggum iterative loops.
|
|
5
|
+
"""
|
|
6
|
+
|
|
7
|
+
from dataclasses import dataclass, field
|
|
8
|
+
from typing import List, Optional
|
|
9
|
+
|
|
10
|
+
|
|
11
|
+
@dataclass
|
|
12
|
+
class RalphConfig:
|
|
13
|
+
"""Configuration for Ralph Wiggum iterative execution"""
|
|
14
|
+
|
|
15
|
+
# Maximum iterations before auto-stop (safety limit)
|
|
16
|
+
max_iterations: int = 10
|
|
17
|
+
|
|
18
|
+
# Promise text to look for to signal completion
|
|
19
|
+
completion_promise: str = "DONE"
|
|
20
|
+
|
|
21
|
+
# Whether to require explicit completion criteria
|
|
22
|
+
require_completion_criteria: bool = True
|
|
23
|
+
|
|
24
|
+
# Minimum confidence score to consider task complete
|
|
25
|
+
min_confidence: float = 0.8
|
|
26
|
+
|
|
27
|
+
# Timeout per iteration in seconds
|
|
28
|
+
iteration_timeout: int = 300
|
|
29
|
+
|
|
30
|
+
# Delay between iterations in seconds
|
|
31
|
+
iteration_delay: float = 1.0
|
|
32
|
+
|
|
33
|
+
# Quality gates to run between iterations
|
|
34
|
+
quality_gates_enabled: bool = True
|
|
35
|
+
|
|
36
|
+
# Stop on first quality gate failure or continue iterating
|
|
37
|
+
stop_on_gate_failure: bool = False
|
|
38
|
+
|
|
39
|
+
# Patterns that indicate completion (besides promise tag)
|
|
40
|
+
completion_patterns: List[str] = field(
|
|
41
|
+
default_factory=lambda: [
|
|
42
|
+
r"<promise>.*?</promise>",
|
|
43
|
+
r"all tests pass",
|
|
44
|
+
r"task complete",
|
|
45
|
+
r"implementation complete",
|
|
46
|
+
]
|
|
47
|
+
)
|
|
48
|
+
|
|
49
|
+
# Patterns that indicate the task is stuck and should stop
|
|
50
|
+
stuck_patterns: List[str] = field(
|
|
51
|
+
default_factory=lambda: [
|
|
52
|
+
r"cannot proceed",
|
|
53
|
+
r"blocked by",
|
|
54
|
+
r"need human intervention",
|
|
55
|
+
r"unable to resolve",
|
|
56
|
+
]
|
|
57
|
+
)
|
|
58
|
+
|
|
59
|
+
def validate(self) -> List[str]:
|
|
60
|
+
"""Validate configuration, return list of errors"""
|
|
61
|
+
errors = []
|
|
62
|
+
|
|
63
|
+
if self.max_iterations < 1:
|
|
64
|
+
errors.append("max_iterations must be at least 1")
|
|
65
|
+
|
|
66
|
+
if self.max_iterations > 100:
|
|
67
|
+
errors.append("max_iterations should not exceed 100 (safety limit)")
|
|
68
|
+
|
|
69
|
+
if self.iteration_timeout < 30:
|
|
70
|
+
errors.append("iteration_timeout must be at least 30 seconds")
|
|
71
|
+
|
|
72
|
+
if not 0.0 <= self.min_confidence <= 1.0:
|
|
73
|
+
errors.append("min_confidence must be between 0.0 and 1.0")
|
|
74
|
+
|
|
75
|
+
return errors
|