Parallel SIW Implementation
Implement multiple SIW issues simultaneously using Agent Teams. Each teammate implements one issue with a full context window, following the kramme:siw:implement-issue workflow.
Arguments: "$ARGUMENTS"
Prerequisites
This skill requires Agent Teams to be enabled (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS). If teams are not available, print:
Agent Teams are not enabled. Use /kramme:siw:implement-issue to implement issues one at a time, or enable teams: Add CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 to settings.json
Then stop.
Workflow
Step 1: Read SIW State
- •Read
siw/OPEN_ISSUES_OVERVIEW.mdto understand all issues and their statuses - •Read the main spec file (from
siw/directory) for project context - •Read
siw/LOG.mdfor current progress and decisions
Step 2: Identify Parallelizable Issues
If specific issue IDs provided (e.g., G-001 P1-002):
- •Validate each issue exists and is in READY status
- •Warn if any have unresolved blockers
If "phase N" provided (e.g., phase 1):
- •Select all READY issues with prefix matching that phase (
P1-*for phase 1)
If no arguments:
- •Select all READY issues across all phases that have no unresolved blockers
Step 3: Analyze File Ownership
For each candidate issue:
- •Read the issue file from
siw/issues/ - •Extract "Affected Areas", "Files to Modify", or "Scope" sections
- •Build a file-to-issue map
Identify conflicts:
- •Issues that touch overlapping files cannot run in parallel
- •Group non-overlapping issues into parallelizable batches
Batching strategy:
- •Batch 1: Maximum set of issues with no file overlaps
- •Batch 2: Issues that were blocked by Batch 1 file conflicts, OR issues whose SIW blockers are resolved by Batch 1 completions
- •Continue until all issues are batched
Step 4: Present Plan
Use AskUserQuestion:
header: "Parallel Implementation Plan"
question: "Ready to implement X issues. Here's the plan:"
options:
- label: "Start parallel implementation"
description: |
Batch 1 (parallel): [issue-ids] - X teammates
Batch 2 (after batch 1): [issue-ids] - Y teammates
Potential conflicts: [details if any]
- label: "Adjust plan"
description: "Let me modify which issues to include"
- label: "Cancel"
description: "Don't implement anything"
Step 5: Spawn Implementation Team
Create a team named siw-implement.
For each issue in Batch 1, spawn a teammate with:
- •Issue content: Full text of the issue file
- •Spec context: Relevant sections of the main spec and any supporting specs
- •File ownership: "You have exclusive write access to: [files]. Do NOT modify files outside this list without messaging the lead first."
- •Workflow: "Follow the
kramme:siw:implement-issueworkflow using the Autonomous Implementation approach:- •Explore codebase for patterns
- •Create technical plan
- •Implement iteratively
- •Run
kramme:verify - •Update the issue file status to DONE
- •Update
siw/OPEN_ISSUES_OVERVIEW.md - •Message the lead when complete"
- •Plan approval: Require plan approval before implementation begins. The lead reviews each teammate's technical plan and approves or rejects with feedback.
Create one task per issue: "Implement [issue-id]: [title]"
Step 6: Monitor Progress
While teammates work:
- •
Review plans: When a teammate submits their implementation plan, review it for:
- •Alignment with spec and issue requirements
- •No file conflicts with other teammates' plans
- •Appropriate patterns and conventions
- •Approve or reject with specific feedback
- •
Track completion: Monitor TaskList for completed tasks
- •
Handle file conflicts: If a teammate discovers it needs a file outside its ownership:
- •Check if the owning teammate is done with that file
- •If yes: grant access
- •If no: queue the request, or suggest the teammate implement a different approach that stays within its files
- •
Handle blockers: If a teammate gets stuck:
- •Provide additional context about the codebase
- •Suggest alternative approaches
- •In worst case, reassign the issue to a different teammate
Step 7: Proceed to Next Batch
When all Batch 1 tasks complete:
- •Update
siw/LOG.mdwith Batch 1 completions - •Check if Batch 2 issues are now unblocked (both SIW dependency and file ownership)
- •Assign Batch 2 issues to idle teammates or spawn new ones
- •Repeat monitoring until all batches complete
Step 8: Final Verification
After all issues are implemented:
- •Run
kramme:verifyon the full scope to check for integration issues - •If verification fails:
- •Identify which teammate's changes caused the failure
- •Either resume that teammate to fix, or fix directly
- •Update
siw/LOG.mdwith session summary:
## Current Progress
**Last Updated:** {date}
**Quick Summary:** Parallel implementation of X issues
### Project Status
- **Status:** In Progress | **Completed this session:** {issue-ids}
### Last Completed
- {issue-id}: {title} (Batch 1)
- {issue-id}: {title} (Batch 1)
- {issue-id}: {title} (Batch 2)
### Decisions Made
- [Any cross-cutting decisions discovered during parallel implementation]
### Next Steps
1. {next ready issue or phase}
- •Check for phase completion (same as
kramme:siw:implement-issueStep 11.2)
Step 9: Spec Sync
Review all decisions logged by teammates. If any need spec updates, follow kramme:siw:implement-issue Step 10 (Sync Decisions to Spec).
Step 10: Cleanup
- •Shut down all teammates
- •Clean up the team
File Conflict Prevention
This skill uses a multi-layer approach:
- •Pre-analysis: Before spawning, the lead reads each issue's affected areas and builds a file ownership map
- •Exclusive ownership: Each teammate gets an explicit list of files it can write to
- •Batching: Issues with file overlaps go into sequential batches
- •Plan approval: Lead reviews each teammate's plan to catch file conflicts early
- •Runtime messaging: Teammates message the lead if they discover they need files outside their set
- •Post-verification:
kramme:verifycatches any integration issues after all implementations
Usage
/kramme:siw:implement-parallel # Implement all READY issues with no blockers /kramme:siw:implement-parallel phase 1 # Implement all READY Phase 1 issues /kramme:siw:implement-parallel P1-001 P1-003 G-002 # Implement specific issues in parallel
When to Use This vs /kramme:siw:implement-issue
Use this skill when:
- •Multiple READY issues exist with no blocking dependencies
- •Issues touch different files/modules
- •You want to speed up a phase by parallelizing independent work
Use /kramme:siw:implement-issue when:
- •Implementing a single issue
- •Issues are tightly coupled and need sequential implementation
- •You want lower token cost