# v1.8 Update
Release Date: 2026-02 (Planned)
## Summary
v1.8 introduces performance optimizations and feature enhancements: improvement cycle log disabling, PRE_COMMIT order change for performance gains, and exploration-only mode (`--only-explore`) for efficient investigation tasks.
**Performance Improvements**:
- v1.7 parallel execution (27-35s reduction): search_text multiple patterns, Read/Grep parallel execution
- v1.8 optimizations (6-9s reduction): Improvement cycle log disabling, PRE_COMMIT order change
- **Total reduction: 33-44 seconds** (approximately 8-11% faster)
## New Features
### 1. Improvement Cycle Log Disabling
The improvement cycle feature (outcomes.jsonl / decisions.jsonl, introduced in v1.0) had the following issues:
1. **Non-functional**:
- `get_session_status` does not return completed sessions
- Step 0 failure detection does not work
- Similarity matching is unimplemented
2. **Performance impact**:
- Step 0: Failure check (2-3s)
- Step 11: `record_outcome` call (2-3s)
- Total: 4-6s wasted
**Changes**:
- `tools/outcome_log.py`: File writes disabled
- `code_intel_server.py`: `record_decision` calls removed
- `.claude/commands/code.md`: Step 0 disabled
**Reduction**: 4-6 seconds (2-3 LLM round-trips)
---
### 2. EXPLORATION Phase Parallelization (v1.7 implementation)
**Note**: This feature was implemented in v1.7 and is available in v1.8 as an existing capability.
`search_text` extended to support multiple patterns for parallel execution:
**Usage**:
```python
# Single pattern
search_text(pattern="modal")
# Multiple patterns (parallel execution)
search_text(patterns=["modal", "background", "overlay"])
```
**Reduction**: 20 seconds (52s → 32s, 38% reduction)
For details, see [v1.7 Update](v1.7.md).
---
### 3. PRE_COMMIT + QUALITY_REVIEW Order Change
**Current implementation**: Quality check after commit:
```
Current flow:
Step 10: PRE_COMMIT
├─ review_changes (garbage detection + keep/discard)
├─ finalize_changes (commit execution) ★Commit here
└─ PRE_COMMIT complete
Step 10.5: QUALITY_REVIEW
├─ Quality check based on quality_review.md
├─ Issues found → Revert to READY (after fix, new commit in PRE_COMMIT)
└─ No issues → merge_to_base
```
**Problems**:
1. **Quality check after commit** → More commits on rework
2. **Git history complexity**: New commit created for each fix
3. **Inefficiency**: One commit would suffice with quality check before commit
**Improved flow**: Quality check before commit:
```
Improved flow:
Step 10: PRE_COMMIT (integrated flow)
├─ review_changes (garbage detection + keep/discard)
├─ Quality check (quality_review.md)
├─ Issues found?
│ YES → Revert to READY
│ NO → finalize_changes (commit execution) ★Commit if no issues
└─ To merge_to_base
```
**Implementation**:
- `tools/session.py`: Added commit preparation state fields
- `tools/branch_manager.py`: Added `execute_commit=False` parameter, `execute_prepared_commit()` method
- `code_intel_server.py`: Modified `finalize_changes` and `submit_quality_review`
**Reduction**: 2-3 seconds (eliminates extra commit round-trip on quality issues)
---
### 4. Exploration-Only Mode (Intent-based + --only-explore)
#### Background
For investigation tasks such as codebase problem analysis or impact assessment without implementation, the following issues existed:
1. **Unnecessary phase execution**:
- READY, POST_IMPL_VERIFY, PRE_COMMIT, QUALITY_REVIEW executed even when only exploration was needed
- No option to skip implementation phases
2. **Gaps in existing options**:
- `--only-verify`: Verification only (skips exploration too)
- `--quick` / `--fast`: Skip exploration (executes implementation)
- No "exploration-only" option
#### Changes
Exploration-only mode is achieved through **two methods**:
##### A. Intent-based Automatic Mode
**Automatic configuration via Intent Classification**:
- Intent=INVESTIGATE or Intent=QUESTION → Automatically sets `skip_implementation=true`, `skip_branch=true`
- No flag required (system auto-detects)
**Usage examples**:
```bash
# These automatically enter exploration-only mode
/code What is the implementation of sample/calculator.py? # Intent=QUESTION
/code Investigate how authentication is implemented # Intent=INVESTIGATE
```
##### B. --only-explore Flag (Explicit Override)
**Exploration-only mode for any Intent**:
- Explicitly specify with `--only-explore` / `-e` flag
- Can force exploration-only even for Intent=IMPLEMENT/MODIFY
**Usage examples**:
```bash
# Investigation only, no implementation
/code --only-explore Add login feature # Intent=IMPLEMENT but exploration only
/code -e Fix this bug # Intent=MODIFY but exploration only
```
**Comparison of the two methods**:
| Method | Trigger | Intent | Branch |
|--------|---------|--------|--------|
| Intent-based | Automatic (INVESTIGATE/QUESTION) | INVESTIGATE/QUESTION | Not created |
| --only-explore | Explicit flag | Any | Not created |
**Executed phases** (common to both):
```
✅ DOCUMENT_RESEARCH
✅ EXPLORATION (find_definitions, find_references, search_text)
✅ Symbol Validation
✅ SEMANTIC (if needed)
✅ VERIFICATION (if needed)
✅ IMPACT_ANALYSIS
✅ Report findings to user
❌ READY (implementation) - Skipped
❌ POST_IMPL_VERIFY - Skipped
❌ PRE_COMMIT - Skipped
❌ QUALITY_REVIEW - Skipped
❌ Branch creation - None (exploration only)
```
**Phase matrix**:
| Option | Explore | Implement | Verify | Intervention | Garbage | Quality | Branch |
|--------|:-------:|:---------:|:------:|:------------:|:-------:|:-------:|:------:|
| Intent=INVESTIGATE/QUESTION | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
| `--only-explore` / `-e` | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
#### Implementation Details
**File changes**:
1. `.claude/commands/code.md`:
- Added INVESTIGATE/QUESTION explanation to Step 1: Intent Classification
- Added Intent-based skip_implementation auto-configuration to Step 2: Session Start
- Added skip_branch determination logic to Step 3.5: begin_phase_gate (skip_implementation=true → skip_branch=true)
- Added `--only-explore` / `-e` processing to Step -1: Flag Check
- Added exploration completion handling to Step 8: IMPACT_ANALYSIS
2. `tools/session.py`:
- Added `skip_implementation: bool` flag to `SessionState`
- Modified `submit_impact_analysis` to return `exploration_complete=true` when `skip_implementation=True`
3. `code_intel_server.py`:
- Added `skip_implementation` parameter to `start_session`
- Added skip_implementation check to `begin_phase_gate` (skip_implementation=true → skip_branch=true, phase=EXPLORATION)
**Branch creation fix**:
- No branch is created in exploration-only mode (skip_implementation=true)
- `begin_phase_gate` automatically sets skip_branch=true
- Reason: No code changes in exploration-only, so branch is unnecessary
**Performance impact**: None (extends functionality rather than reducing time)
**Use cases**:
- Codebase problem investigation (Intent=INVESTIGATE auto-detects)
- Security audit preliminary research (Intent=INVESTIGATE auto-detects)
- Architecture understanding (Intent=QUESTION auto-detects)
- Impact analysis (Intent=INVESTIGATE auto-detects, or --only-explore explicit)
---
## Future Optimization Direction
### v1.9 and beyond (further improvements)
- sync_index acceleration (Embedding batch processing) → Planned for [v1.9](v1.9.md)
- VERIFICATION + IMPACT_ANALYSIS integration (10s reduction potential) → Planned for [v1.9](v1.9.md)
- DOCUMENT_RESEARCH selective execution (planned separately)
### Long-term
- Reduce inter-phase wait times
- LLM thinking efficiency (simpler flow design)
---
## Breaking Changes
- None
## Bug Fixes
- None
---
## Related Changes
- `.claude/commands/code.md`: Flag definition and phase flow updated
- `tools/session.py`: Session state extended with skip_implementation flag
- `code_intel_server.py`: Session creation parameter extended
- `README.md` / `README_ja.md`: Documentation updated with new option
