Pipeline
Run the complete feature pipeline (align -> implement -> review) in fresh sessions
Claude CodeGeneric
---
description: "Run the complete feature pipeline (align -> implement -> review) in fresh sessions"
argument-hint: "<feature-id> [new|update]"
---
# /pipeline - Multi-Session Feature Pipeline
Orchestrates feature completion across fresh sessions to maintain quality.
## Usage
/pipeline f1 # New feature implementation /pipeline f2 update # Update existing feature
## What It Does
1. **Phase 1: Alignment** (fresh session)
- Runs `/align` command for spec verification
- Outputs alignment report to `.claude/handoff/`
- Commits docs if ready
- STOPS if blockers found
2. **Phase 2: Implementation** (fresh session, with iteration)
- Reads alignment report as input
- TDD workflow: tests first, then implementation
- **Build verification** after each attempt
- **Test verification** after successful build
- **Retry loop** (max 3 attempts) if build/tests fail
- Only commits after build + tests pass
3. **Phase 3: Review** (fresh session)
- Runs `/review` command
- Verifies implementation matches specs
- Outputs review report
- Reports PASS/FAIL
## Configuration
| Setting | Default | Description |
|---------|---------|-------------|
| `MAX_IMPL_ATTEMPTS` | 3 | Max retry attempts for implementation |
| `XCODE_DESTINATION` | iPhone 17 | Simulator for build/test |
## Run Script
```bash
bash .claude/scripts/feature-pipeline.sh $ARGUMENTS
Pipeline Flow
Phase 1: Alignment
│
├── /align <feature>
├── Check: Ready to Implement?
│ ├── NO → EXIT (fix blockers)
│ └── YES → Continue
└── Commit docs
│
▼
Phase 2: Implementation (with iteration)
│
├── Attempt 1/3
│ ├── Claude implements with TDD
│ ├── xcodebuild build
│ │ └── FAIL → Attempt 2/3
│ └── xcodebuild test
│ └── FAIL → Attempt 2/3
│
├── Attempt 2/3 (if needed)
│ ├── Claude reads errors, fixes
│ ├── xcodebuild build
│ └── xcodebuild test
│
├── Attempt 3/3 (if needed)
│ └── Same as above
│
├── All attempts failed? → EXIT
└── Commit code (only if build+tests pass)
│
▼
Phase 3: Review
│
├── /review <feature>
├── Check: Review Result?
│ ├── FAIL → EXIT (issues to fix)
│ └── PASS → Continue
└── Pipeline Complete ✓
Why Fresh Sessions?
Each phase starts with clean context:
- No accumulated biases from previous work
- Forces reliance on documentation
- Validates handoff document quality
- Prevents quality degradation
Why Iteration in Phase 2?
Inspired by Ralph Wiggum technique:
- Build/test failures are common on first attempt
- Retry with error context improves success rate
- Max 3 attempts prevents infinite loops
- Only commits verified working code
Maintain Pipeline?
Let people know it's listed here — add the badge (live metrics, light/dark aware) or a plain link to your README or docs.
[Pipeline on getagentictools](https://getagentictools.com/loops/anh-ht93-pipeline-multi-session-feature-pipeline?ref=badge) npx agentictools info loops/anh-ht93-pipeline-multi-session-feature-pipeline The second line is the CLI lookup for this page — handy in READMEs and docs.