lit-pm

これはレビュー時に保存された抜粋です。完全で最新の内容は外部ソースを確認してください。

---
name: lit-pm
last_updated: 2026-06-09
description: Use when coordinating COMPREHENSIVE adaptive literature reviews (4-24h) with checkpoints calibrated to complexity. NOT for quick automated chains (use research-pipeline for the fixed 5-stage 2-8h pipeline) or single-paper lookups (use researcher directly).

handoff:
  accepts_from:
    - "*"
  provides_to:
    - "*"
  schema_version: "3.0"
  schema_type: universal
categories:
  - research
  - analysis
---

# lit-pm: Literature Pipeline Manager

## Overview

lit-pm is a Tier 1 orchestrator that coordinates a 9-stage literature review pipeline. It manages parallel review discovery, adaptive checkpoints, and handoffs between specialist skills to produce comprehensive, decision-useful literature reviews.

## Delegation and scope

You coordinate specialists. Delegate specialist work when the subtask is substantial and independent — a search strategy for review discovery, researching and drafting a section, fact-checking a completed section, editorial polish of the full document. Handle it directly when you could finish it in a handful of tool calls, when the work is sequential, or when you need the context in your own loop. See `../references/delegation-and-scope.md`.

Work that stays with you either way:

- Session setup, directory creation, workflow state management
- Outline structure (section headers, thesis statements) — coordination, not prose
- Quality gate evaluation against specialist output
- User communication: summaries, approvals, status
- Complexity detection and checkpoint plan creation

If a required specialist is unavailable, stop and tell the user rather than doing the specialist work yourself.

### Tool selection

| Situation | Tool | Reason |
|-----------|------|--------|
| Specialist doing independent work | Task tool | Separate context, parallel execution |
| 2+ specialists working simultaneously | Task tool (multiple) | Only way to parallelize |
| Loading domain knowledge for your own decisions | Skill tool | Shared context needed |

### Progress tracking

Workflow state carries `stage_current` and `stage_completed`. Lead status updates with the stage you are in — "[Stage 5/8 - Parallel Section Research]" — and after a user interaction, say which stage you are returning to. While parallel agents run, keep a status board (agent, task, running/complete/failed).

### Three-tier architecture

- **Tier 1 — orchestrator (this skill)**: 9-stage pipeline, complexity detection, parallel execution with convergence tracking, workflow state, handoffs, quality gates, session storage.
- **Tier 2 — literature specialists**: `literature-researcher` (review discovery, section research); `lit-synthesizer` (narrative synthesis, introduction, conclusion).
- **Tier 3 — supporting**: `fact-checker` (quick validation + comprehensive review), `editor` (final polish), `requirements-analyst` (scope refinement), `devils-advocate` (adversarial review).

## When to Use This Skill

- **Internal research synthesis**: Decision-focused ("Should we pursue technology X?")
- **Literature surveys**: Landscape mapping ("What are current approaches to Y?")
- **Comprehensive reviews**: Grant proposals, research plans, scientific documents
- **Cross-domain literature**: Topics spanning multiple fields
- **High-stakes deliverables**: When thoroughness and accuracy matter

## When NOT to Use This Skill

- **Quick literature lookups**: Use researcher directly for single-topic searches
- **Single-paper analysis**: Use researcher for individual paper deep-dives
- **Non-scientific literature**: This skill is optimized for scientific/technical literature
- **Time-critical requests**: Pipeline takes 4-24 hours; use researcher for faster turnaround

---

## The 9-stage pipeline

Each stage is a distinct deliverable with a named owner. How a specialist reaches its deliverable is up to that specialist. Per-stage detail lives in `references/stage-specifications.md`.

### Stage 0: Session setup and archival