/backlog

description: Backlog review and cleanup - multi-agent backlog review, quick no-agent pulse (--fast), or interactive triage cleanup
argument-hint: "[review|cleanup] [--fast] [--workflow]"

Backlog

Backlog management in three modes. Review (heavy) launches an explorer + parallel domain agents for multi-perspective backlog insights. `--fast` is a quick no-agent backlog pulse (orchestrator-only, the lightweight sibling of Review). Cleanup is fast interactive triage - no agents.

**Quick no-agent backlog pulse: `/lets:backlog --fast`. For HOW to build something, use `/lets:plan`. For an open question or decision about the project, use `/lets:opinion`.**

> **IMPORTANT:** If the spec below invokes any deferred tool (e.g. `AskUserQuestion`), you MUST load and call it as specified. Never skip the call, never substitute a default answer of your own — the tool invocation is part of the contract. This is critical.

Usage

/lets:backlog            # Ask which mode, then go
/lets:backlog review     # Multi-agent backlog review (skip menu)
/lets:backlog cleanup    # Interactive triage cleanup (skip menu)
/lets:backlog review --workflow  # Run the Review fan-out + aggregate off-context via a Dynamic Workflow

Step 0: Choose Mode

**Parse the argument first:** strip a `--fast` token (sets fast mode - no agents) and a `--workflow` token (sets workflow mode - Review only) if present; the remaining token is the mode keyword. Precedence/compat:

• `--fast` selects the orchestrator-only **Fast mode** (a quick no-agent pulse) regardless of the `review`/no-arg keyword. `--fast` + `--workflow` -> `--fast` wins (orchestrator-only beats off-context fan-out); note it in one line.
• `--fast` passed with `cleanup` is ignored with a one-line note (Cleanup is already no-agent triage).
• `--workflow` passed with `cleanup` is ignored with a one-line note (Cleanup has no agents).

If an argument is provided, parse it: `--fast` -> Fast mode (below); `review` -> Review Backlog Mode; `cleanup` -> Cleanup Mode. A remaining token that is **neither** `review` nor `cleanup` is an **area/topic filter** -> Fast mode with that token as the F1 filter (e.g. `/lets:backlog --fast auth` scopes the pulse to auth; a bare `/lets:backlog <area>` implies Fast mode too).

If no argument, use AskUserQuestion:

AskUserQuestion(
  questions=[{
    question: "What do you want to do with the backlog?",
    header: "Backlog",
    options: [
      { label: "Review backlog", description: "Agents analyze project state, generate ideas and surface gaps (~2-3 min)" },
      { label: "Quick pulse", description: "No agents - fast orchestrator-only scan and conversation" },
      { label: "Cleanup", description: "Triage stale tasks - close, merge, reprioritize" }
    ],
    multiSelect: false
  }]
)

**Handle response:**

• **Review backlog** -> Review Backlog Mode (below)
• **Quick pulse** -> Fast mode (below)
• **Cleanup** -> Cleanup Mode (below)

---

Fast Mode (--fast)

No agents. The orchestrator gathers backlog context directly and enters conversation - a quick pulse, the lightweight sibling of the agent-backed Review below (same house `--fast` convention as `/lets:plan --fast`: no flag = subagents, `--fast` = orchestrator-only).

Step F1: Gather Context

Start with the shared snapshot so the pulse builds on the same "where are we" view as `/lets:status`:

Invoke `Skill(skill: "lets:orient")`.

Then gather the backlog-specific extras the snapshot doesn't cover:

LETS_PROJECT_ROOT=$(git rev-parse --show-toplevel)

list-by-status status=open limit=30

(`stats` + the in_progress list already came from the orient snapshot above - don't re-fetch them.)

git log --oneline -15

Also use the Grep tool (not bash grep) to scan for tech-debt signals:

Grep(pattern="TODO|FIXME|HACK|XXX", path="{LETS_PROJECT_ROOT}", output_mode="content", head_limit=20)

If an area/epic argument was passed alongside `--fast`:

• the tracker's `search` verb on {argument} (if no results, fall back to `list-by-status` and scan titles)
• Grep/Glob for related code

Step F2: Open with Observations

Based on the orient snapshot above plus the backlog signals gathered, present 3-5 proactive observations:

## Backlog Pulse

Based on what I see in the project:

1. **{observation}** - {why it matters, 1-2 sentences}
2. **{observation}** - {why it matters}
3. **{observation}** - {why it matters}

**Question:** {probing question that opens discussion}

Categories to draw observations from:

• Gaps: areas with activity but no tasks
• Stale work: tasks open long with no progress
• Patterns: recurring themes across tasks/commits
• Quick wins: small improvements with high impact
• Risks: things that could break or become debt

Step F3: Interactive Dialog

After each user response: 1. **Acknowledge** - brief, no fluff 2. **Build** - add insight or connection 3. **Probe** - ask next question that goes deeper

If the user wants the heavier multi-agent pass: "Want a deeper backlog review? `/lets:backlog review` launches expert agents."

Step F4: Capture & Exit

If active task:

comment-add task=<task-id> body="Backlog pulse: {key takeaways, 2-3 items}"

If ideas emerged that deserve tasks, offer to create them (use the `create-task` skill, user approval per task).

---

Review Backlog Mode

The one heavy mode: explorer scouts the project, parallel domain agents ideate over the backlog, results aggregate.

Execution path (standard vs workflow)

The scout (Phase 1) and agent selection (Phase 2) ALWAYS run in-context. Only Phases 3-4 (fan-out + aggregate) can move off-context via a Dynamic Workflow - `--workflow` is a transparent perf lever, same findings.

• If `--workflow` was parsed -> use the workflow path (go to `## Workflow Mode` after Phase 2). An explicit `--workflow` always wins.
• Else if the `Workflow` tool is **not** available this session -> silently use th