Initiative Review

Overview

Validate and fix initiative tasks before execution using specialized auditor subagents.

Core principle: Each task executes in complete isolation. If an agent can't succeed with only the task description, initiative context, and design doc — the task will fail.

Announce at start: "Running initiative review with 3 parallel auditors..."

The Process

Step 1: Gather Context

orc initiative show INIT-XXX

Extract:

• Initiative ID
• Design doc path (from vision/description)
• Task count and dependency structure

Step 2: Invoke Auditor Subagents

Launch these three auditors in parallel using Task tool:

Each auditor is defined in .claude/agents/ and knows its output format.

Prompt template for each:

Audit INIT-{ID} for {coverage/dependencies/description quality}.
Design doc: {path}
Return structured findings with specific fix recommendations.

Step 3: Consolidate Findings

After auditors complete, synthesize into action plan:

## Critical Issues (Block Execution)
| Issue | Impact | Fix |
|-------|--------|-----|

## Should Fix (Quality)
| Issue | Tasks | Improvement |
|-------|-------|-------------|

## Action Items
1. [ ] Create missing tasks
2. [ ] Add/remove dependencies
3. [ ] Update task descriptions

Step 4: Execute Fixes

With user approval, execute the fixes. The mechanics depend on what's broken:

Creating Missing Tasks

orc new "Title" -w medium -c feature -i INIT-XXX -d "Description..."

Look at existing tasks in the initiative for description patterns. Match the style.

Fixing Dependencies

# Add missing dependency
orc edit TASK-X --add-blocker TASK-Y

# Remove unnecessary dependency
orc edit TASK-X --remove-blocker TASK-Y

# Cross-initiative dependency
orc edit TASK-X --add-blocker TASK-Y  # Y from other initiative

Updating Descriptions

orc edit TASK-X -d "Updated description..."

When updating descriptions, preserve what's good and add what's missing:

• Keep existing design doc references
• Add file paths if missing
• Add acceptance criteria if missing
• Add scope boundaries if missing

Step 5: Re-Run Auditors

After fixes, re-run the auditors to verify. Continue until:

• Coverage: No gaps, no orphans
• Dependencies: No missing, no over-constraints
• Descriptions: All tasks rated "Good"

Flexibility Notes

Projects differ. The auditors return what's wrong and recommendations, but:

• File paths: Auditors suggest paths based on project structure they observe
• Description style: Match existing task descriptions in the initiative
• Dependency logic: Use the "would tests fail?" test, not intuition
• Fix commands: Use orc edit --help and orc new --help for current flags

The skill teaches the process. The auditors analyze the specific project. You execute fixes appropriate to the context.

Quick Reference

Red Flags (Stop and Fix)

• Cross-initiative dependency mentioned but not set
• 10+ tasks with zero dependencies (suspicious)
• Task says "see design doc" without path
• Large task touching 2 files (wrong weight)
• Circular dependencies
• Task says "deprecate" or "either X or Y" (hedging - pick one)
• Design says "replace X" but task doesn't say DELETE X
• Major decision left to executing agent ("choose between...")