consolidate

name	consolidate
description	Use when the backlog feels unwieldy — after a blossom run, before a sprint, or when you suspect duplicates or stale tasks. Deduplicates, fills gaps, and cleans up dependencies. Keywords: backlog, cleanup, dedup, triage, hygiene, review, organize.
argument-hint	[area or scope]
disable-model-invocation	false
user-invocable	true
allowed-tools	Read, Grep, Glob, Bash(git:*)
context	fork

Consolidate: Backlog Review and Tightening

You are running the Consolidate workflow — a structured pass over the current task backlog to eliminate sprawl, fill gaps, and sharpen priorities. Target area (optional): $ARGUMENTS

Review the current backlog using your project's task tracking approach. If only a TODO.md file exists, review it manually: scan for duplicates, reorder by priority, and remove items already completed (check git log).

When to Use

• After a blossom run produced a large backlog
• When the backlog has accumulated over multiple sessions and feels unwieldy
• Before starting a new sprint of work — get a clean picture first
• When you suspect duplicate or stale tasks exist

Overview

Consolidate works in 6 phases:

Survey [gather]
  -> Dedup [filter: not duplicate]
    -> Vertical slice audit [assess: complete/missing-companion]
      -> Stale detection [filter: not stale]
        -> Dependency cleanup [verify]
          -> Report [pipe format]

---

Phase 1: Survey (/gather pattern — collect backlog info with sources)

1a. Get the Full Picture

Survey all open tasks, in-progress work, and blocked items using your project's task tracking approach.

1b. Build a Mental Map

For each open task, note:

• Scope: What code/area does this touch?
• Status: Is anyone working on it? Is it blocked?
• Staleness: When was it last updated? Is it still relevant?
• Connections: What does it block or depend on?

If $ARGUMENTS specifies a target area (e.g., "sandbox", "frontend"), focus only on tasks related to that area. Otherwise, review the full backlog.

1c. Categorize

Group tasks into clusters by the area they touch (e.g., "domain/sandbox", "infrastructure", "CLI", "frontend"). This grouping drives the rest of the workflow.

---

Phase 2: Dedup (/filter pattern — criterion: "not duplicate")

2a. Find Overlaps

Within each cluster, look for:

• Exact duplicates: Same work described differently
• Subset tasks: One task is entirely contained within another
• Convergent tasks: Different approaches to the same problem

2b. Merge Strategy

For each overlap found:

1. Exact duplicates: Close the newer one with a note pointing to the older one
2. Subset tasks: Close the subset, add its specifics as notes on the parent
3. Convergent tasks: Keep the better-scoped one, close the other

For each closure, record the reason (e.g., "Duplicate of [other task]", "Absorbed into [parent task]", "Superseded by [better task]").

2c. Report Dedup Results

Track: how many tasks closed, how many merged, how many clusters remain.

---

Phase 3: Vertical Slice Audit (/assess pattern — rubric: complete / missing-companion)

3a. Discover the Project's Architecture

Before auditing slices, identify the project's actual architectural layers. Read the project structure, CLAUDE.md, and top-level directories to determine the pattern in use. Common patterns:

Architecture	Typical Layers
DDD / Hexagonal	Domain, Infrastructure, Application, Interface
MVC	Model, View, Controller, Service
Component-based	Feature modules (auth, billing, etc.) with internal structure
Flat / scripts	Modules, utilities, entry points
Frontend	Components, State/Store, API layer, Routing

Record the discovered layers before proceeding. If the architecture is unclear, list the top-level directories and use those as the layer categories.

3b. Check Each Task for Slice Completeness

For every task that touches a core/inner layer, verify companion tasks exist across the relevant architectural boundaries. The specific companions depend on the architecture discovered in 3a:

• New core logic → Does a data access / persistence task exist?
• New data access → Does a wiring / bootstrapping task exist?
• New service / use case → Does at least one exposure point (CLI, API, UI) exist?
• New events / messages → Does at least one consumer task exist?
• New exposure point → Does the underlying logic task exist?
• Any new feature → Does a test task exist?

3c. Fill Gaps

For each missing companion, create as a child of the epic (if one exists) or standalone:

Create each missing companion task in your project's task tracker as a child of the epic (if one exists). If using TODO.md, add missing companion tasks under the relevant section.

3d. Wire Dependencies

Ensure the natural flow follows the project's dependency direction (inner layers before outer layers). General pattern:

• Core/model tasks block data-access/infrastructure tasks
• Data-access tasks block service/application tasks
• Service tasks block interface/exposure tasks
• Test tasks can run in parallel with their layer

Wire dependencies so outer layers depend on inner layers in your task tracker.

---

Phase 4: Stale Detection (/filter pattern — criterion: "not stale")

4a. Identify Stale Tasks

A task is potentially stale if:

• It was created more than 2 weeks ago with no progress
• Its parent epic has been closed or abandoned
• The code area it targets has been significantly refactored since creation
• It describes work that has already been done (check git log)

4b. Verify Before Closing

For each potentially stale task:

1. Check git log — has this work already been done?

   git log --oneline --all --grep="<relevant keywords>"
   

2. Check the code — does the issue still exist?
3. Check dependencies — would closing this unblock anything?

4c. Close Stale Tasks

Close stale tasks with a reason (e.g., "Stale: [explanation]" or "Already completed in commit [hash]").

---

Phase 5: Dependency Cleanup (/verify pattern — check dependency validity)

5a. Remove Redundant Transitive Dependencies

If A depends on B, and B depends on C, then A does NOT need a direct dependency on C. The transitive dependency through B is sufficient. Review each task's dependency list and remove redundant transitive edges.

5b. Check for Cycles and Validate Structure

Check for dependency cycles in your task tracker. If cycles are found, break them by identifying which dependency is the weakest (most optional) and removing it. Validate the full epic structure for orphaned or disconnected tasks and re-parent or wire them.

5c. Validate Epic Structure

Every epic should have its tasks as children, not just blocking deps. Verify all child tasks are properly parented and that epic status reflects completion progress. Flag any orphaned tasks for re-parenting.

---

Phase 6: Report (pipe format output)

6a. Summarize Changes

Emit the consolidation report in pipe format:

## Consolidated Backlog for [area or "full backlog"]

**Source**: /consolidate
**Input**: [target area or "full backlog review"]

### Items

1. **Closed (dedup)** — X tasks merged or deduplicated
   - source: Phase 2
2. **Closed (stale)** — X tasks no longer relevant
   - source: Phase 4
3. **Created (gap fill)** — X companion tasks added for vertical slices
   - source: Phase 3
4. **Dependencies cleaned** — X redundant deps removed, X new deps added
   - source: Phase 5

### Backlog Health

| Metric | Before | After |
|--------|--------|-------|
| Open tasks | X | X |
| Blocked | Y | Y |
| In-progress | Z | Z |
| Clusters | [list] | [list] |

### Recommendations

**Sharpening gate:** Each recommendation must pass three tests:
1. Name the specific bead/area/cluster
2. State what concretely should change
3. Make it actionable (a single command or skill invocation)

**Before:** "Some clusters might benefit from prioritization review"
**After:** "Reprioritize tasks abc1, def2, ghi3 in the auth cluster from P3→P2 — they block 4 downstream tasks"

- [Recommendation 1: specific action]
- [Recommendation 2: specific action]
- [Recommendation 3: specific action]

### Summary

[One paragraph synthesis of the consolidation: what changed, current backlog health, and recommended next steps.]

---

Guidelines

• Be conservative with closures — when in doubt, keep a task open and add a note
• Check the code before declaring something stale — read, don't guess
• Preserve context — when closing or merging, always explain why in the reason
• Don't reprioritize aggressively — small adjustments only, unless the user directs otherwise
• Focus on the target area if $ARGUMENTS was provided — don't let scope creep into unrelated clusters
• Tool-agnostic — use whatever task tracking approach the project has configured; adapt commands to the available tooling

See also: /gather, /filter, /assess, /verify, /rank, /sprint