@exaudeus/workrail 0.6.1-beta.11 → 0.6.1-beta.12
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.
package/package.json
CHANGED
|
@@ -1,5 +1,67 @@
|
|
|
1
1
|
# Changelog - Systematic Bug Investigation Workflow
|
|
2
2
|
|
|
3
|
+
## [1.1.0-beta.6] - 2025-11-06
|
|
4
|
+
|
|
5
|
+
### 🎯 Major Enhancements: Concrete Instrumentation & Breadth Check
|
|
6
|
+
|
|
7
|
+
**Problem**: Agents were stopping mid-workflow to ask for permission (even in high auto mode), and instrumentation/evidence collection steps were too vague, causing confusion about what to actually do.
|
|
8
|
+
|
|
9
|
+
**Solutions Applied**:
|
|
10
|
+
|
|
11
|
+
#### 1. **New Phase 1f: Final Breadth & Scope Verification**
|
|
12
|
+
Added critical checkpoint AFTER code analysis (Phase 1) but BEFORE hypothesis development (Phase 2):
|
|
13
|
+
- **Catches tunnel vision**: Forces evaluation of 2-3 alternative investigation scopes
|
|
14
|
+
- **Scope sanity check**: Verifies module root, entry points, and component coverage
|
|
15
|
+
- **Wide-angle review**: Ensures sufficient breadth AND depth before committing to hypotheses
|
|
16
|
+
- **Research-backed**: 60% of failed investigations look in wrong place or too narrowly
|
|
17
|
+
- **Decision options**: Continue (scope correct), Expand (add areas), or Shift (wrong place entirely)
|
|
18
|
+
|
|
19
|
+
#### 2. **Phase 3: Detailed Instrumentation Instructions**
|
|
20
|
+
Completely rewrote with concrete, step-by-step guidance:
|
|
21
|
+
- **STEP 1-2**: Review Phase 2g plan, read files needing instrumentation
|
|
22
|
+
- **STEP 3**: Language-specific logging examples (JS/TS, Python, Java)
|
|
23
|
+
- **STEP 4**: Real `search_replace` example with actual code
|
|
24
|
+
- **STEP 5**: File-by-file workflow (read → locate → instrument → verify)
|
|
25
|
+
- **STEP 6**: Fallback for agents without file editing (provide code to user)
|
|
26
|
+
- **AUTO-EXECUTE reinforced**: "DO NOT ask 'Would you like me to continue?'"
|
|
27
|
+
|
|
28
|
+
#### 3. **Phase 4: Detailed Evidence Collection Instructions**
|
|
29
|
+
Added concrete decision tree and execution guidance:
|
|
30
|
+
- **Decision tree**: Can agent run code? → Option A (execute) vs Option B (instruct user)
|
|
31
|
+
- **STEP 1-3**: How to run code, capture logs, verify quality
|
|
32
|
+
- **STEP 4-5**: Parse logs by hypothesis, assess evidence quality
|
|
33
|
+
- **STEP 6**: Complete user instructions template if agent can't execute
|
|
34
|
+
- **STEP 7**: Document evidence with quality scores
|
|
35
|
+
- **AUTO-EXECUTE reinforced**: Ask for SPECIFIC input (how to run tests), not permission to continue
|
|
36
|
+
|
|
37
|
+
#### 4. **metaGuidance: High Auto Mode Clarification**
|
|
38
|
+
Added explicit section:
|
|
39
|
+
> "In HIGH automation mode, agents must execute phases WITHOUT asking for permission between phases. The ONLY confirmations allowed are: (1) Phase 0e early termination, (2) Phase 4a controlled experiments. All other phases execute automatically."
|
|
40
|
+
|
|
41
|
+
### 📊 Changes Summary
|
|
42
|
+
|
|
43
|
+
- **Total steps**: 26 → 27 (added Phase 1f)
|
|
44
|
+
- **Step references**: Updated all from "26 steps" to "27 steps"
|
|
45
|
+
- **Phase 3 prompt**: 2,588 chars → detailed 6-step process with examples
|
|
46
|
+
- **Phase 4 prompt**: Vague execution → detailed decision tree with 7 steps
|
|
47
|
+
- **New checkpoint**: Phase 1f catches wrong-place investigations early
|
|
48
|
+
|
|
49
|
+
### 🎯 Why This Matters
|
|
50
|
+
|
|
51
|
+
**Before**: Agents would:
|
|
52
|
+
- Jump straight from code analysis to hypotheses (tunnel vision)
|
|
53
|
+
- Get confused at Phase 3 ("add instrumentation" - but HOW?)
|
|
54
|
+
- Ask permission between every phase (even in high auto mode)
|
|
55
|
+
- Not know if they should run code or instruct the user
|
|
56
|
+
|
|
57
|
+
**After**: Agents:
|
|
58
|
+
- Verify scope at Phase 1f before committing to hypotheses
|
|
59
|
+
- Have concrete steps: read files, use `search_replace`, examples for each language
|
|
60
|
+
- Execute phases automatically without asking permission
|
|
61
|
+
- Clear decision tree for execution vs user instruction
|
|
62
|
+
|
|
63
|
+
---
|
|
64
|
+
|
|
3
65
|
## [1.1.0-beta.5] - 2025-11-06
|
|
4
66
|
|
|
5
67
|
### 🎯 Major Enhancement: Moved Early Termination Checkpoint to Phase 0e
|