lit-pm
Use when coordinating comprehensive literature reviews requiring multi-stage pipeline (archival setup, scope refinement, parallel review discovery, outline synthesis, section writing, fact-checking, editorial polish). Orchestrates literature-researcher, lit-synthesizer, fact-checker, and editor skills with adaptive checkpoints based on complexity and stakes.
npx skills add majiayu000/claude-skill-registry --skill lit-pm --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
What it does
lit-pm is a Tier 1 orchestrator that coordinates a nine-stage literature review pipeline. It manages parallel discovery, adaptive checkpoints based on complexity and stakes, and handoffs between specialized skills (literature-researcher, lit-synthesizer, fact-checker, editor) with session-based intermediate storage and quality gates.
How it works
- It defines a 9-stage workflow (Stage 0 through Stage 8) with explicit owners for each stage and predefined outputs, checkpoints, and durations.
- It validates the presence of required sub-skills before starting; ABORT occurs if any required skill is missing.
- It implements adaptive orchestration using complexity and stake signals to decide whether stages run automatically, require user approval, or always execute as quality gates.
- It coordinates parallel execution (e.g., Stage 2 discovery uses 2-3 literature-researcher agents; Stage 5 section writing runs in parallel with a queue) and tracks convergence across agents.
- It defines a session-based workflow: creates a temporary session directory, stores archival guidelines, propagates session context through handoffs, and performs session cleanup upon completion or failure.
- It enforces a series of quality gates (both automated and human judgment) at designated stages (e.g., 6a quick validation, 6b comprehensive checks, 6c devil's advocate, 7 synthesis reviews, 8 polish).
- It provides a detailed archival guideline extraction from CLAUDE.md, with a specific YAML/Markdown output path and a fixed archival summary format to be shared with downstream agents.
When to use it
- For internal research synthesis requiring a structured, multi-stage pipeline.
- For literature surveys and comprehensive reviews including grant proposals or scientific documents.
- When cross-domain literature and high stakes demand thoroughness, validation, and editorial polish.
What it can touch
- Coordinates with the following skills: requirements-analyst, literature-researcher, lit-synthesizer, fact-checker, editor, devils-advocate.
- Requires these skills to exist prior to starting Stage 0; on missing skill, it aborts with a clear error.
Caveats
- Pipeline duration is long (Stage 5 can run 3-5 hours per section; overall process up to 4-24 hours as described).
- Timeouts exist per stage with escalation rules (e.g., archival stage has a hard 5-minute timeout).
- Time constraints and complexity detection influence checkpointing and human-in-the-loop requirements.
- Resource limits cap concurrent agents and parallel writers to manage throughput and convergence tracking.
# lit-pm: Literature Pipeline Manager ## Overview lit-pm is a Tier 1 orchestrator skill 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. ### Three-Tier Architecture **Tier 1: Orchestrator (this skill)** - Coordinates 9-stage pipeline (Stage 0-8) - Implements adaptive orchestration (complexity detection -> checkpoint plan) - Manages parallel execution with convergence tracking - Handles workflow state, handoffs, and quality gates - Manages session-based intermediate file storage **Tier 2: Specialized Literature Skills** - `literature-researcher`: Review discovery, section research (15-30 papers per section) - `lit-synthesizer`: Senior scientific author, narrative synthesis, introduction/conclusion **Tier 3: Supporting Skills** - `fact-checker`: Quick validation + comprehensive review - `editor`: Final polish - `requirements-analyst`: Scope refinement ## When to Use This Skill - **Internal research synthesis**: Decision-focused ("Should we pursue technology X?") - **Literature surveys**: Landscape mapping ("What are cur
- Overview
- Three-Tier Architecture
- When to Use This Skill
- When NOT to Use This Skill
- Pre-Flight Validation
- The 9-Stage Pipeline
- Stage 0: Archival Guidelines Review
- Stage 1: Scope Refinement
- Stage 2: Parallel Review Discovery
- Stage 3: Layered Outline Synthesis
- Stage 4: Introduction Writing
- Stage 5: Parallel Section Research & Writing
- Stage 6a: Per-Section Quick Validation (BLOCKING)
- Stage 6b: Comprehensive Fact-Check (NON-BLOCKING)
Standard invocation lit-pm "Comprehensive review of [topic] for [purpose]" With explicit checkpoint control lit-pm --review-outline "Survey of current approaches to [topic]" lit-pm --full-auto "Quick survey of [topic]" Resume interrupted workflow lit-pm --resume workflow-id
What does the lit-pm skill do?
Use when coordinating comprehensive literature reviews requiring multi-stage pipeline (archival setup, scope refinement, parallel review discovery, outline synthesis, section writing, fact-checking, editorial polish). Orchestrates literature-researcher, lit-synthesizer, fact-checker, and editor skills with adaptive checkpoints based on complexity and stakes.
How do I install it?
Run `npx skills add majiayu000/claude-skill-registry --skill lit-pm --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From majiayu000/claude-skill-registry, a repository with 534 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.