| name | autor-overview |
| description | Understand the autor codebase at a high level. Use when the user asks how to use autor, what built-in skills it has, what features it supports, which workflow to choose, or wants a project overview before using other skills. |
autor Overview
When the user has not yet given a concrete task and is asking things like "How do I use this software?", "What skills are available?", "What other features does it have?", or "Where should I start?", use this skill first to build a high-level view, then switch to a more specific skill if needed.
The three things you need to explain
1. How to use autor
Start with the three main entry points:
- Agent mode: launch Claude Code, Copilot, Codex, Cline, or another agent in the project directory, then use natural language for search, reading, writing, and analysis
- CLI mode: run
autor --help directly; ideal for scripting and quick lookups
- MCP mode: run
autor-mcp for Claude Desktop, Cursor, and other MCP clients
When the user asks "How do I get started?", give them the shortest path:
git clone https://github.com/ZimoLiao/autor.git && cd autor
pip install -e ".[full]"
cp config.local.example.yaml config.local.yaml
autor setup check --lang en
Then add the key notes:
- an LLM key is optional; without it, metadata extraction falls back to regex only
- a MinerU key is optional; without it,
.md files can still be processed directly, but cloud PDF -> Markdown parsing is unavailable
- common entry commands include
autor search, autor show, autor pipeline, autor ws, and autor-mcp
2. Which skills autor includes
Summarize them by group. Start with the overview, then go deeper only if the user asks.
Project overview
autor-overview: explains how to use the software, outlines the available skills, clarifies feature boundaries, and recommends where to start
External acquisition / Records service integration
autodownload: explains how autor works with the Records-backed AutoDownload service, where the REST boundary is, which endpoints to use, and how to build or expand a workspace through external acquisition
Knowledge-base management
search: search the local literature library
show: view paper content layer by layer
enrich: extract TOC / generate L3 conclusion cards / backfill abstracts / refresh citation counts
ingest: batch-ingest papers, theses, and general documents
index: rebuild the node-level FTS5 evidence index
explore: fetch independent exploration datasets from OpenAlex
graph: citation-graph and shared-reference analysis
citations: top-cited rankings and citation-count refresh
export: export BibTeX
import: import Endnote / Zotero libraries
rename: normalize paper-directory naming
audit: audit and repair metadata problems
trials: retrieve and organize clinical-trial information
Academic writing
plan: prepare a review in a Springer Nature Reviews style through outline revision, paper classification, and fixed task / table design
literature-review: write a literature review
paper-writing: draft individual paper sections
update: revise a manuscript under reviewer or editor comments with minimum-necessary, traceable edits and evidence-gated outside supplementation
citation-check: verify whether citations are real and accurate
polish: polish academic text, remove AI/workflow artifacts, normalize terminology, and adapt the style
review-response: draft reviewer responses
research-gap: identify research gaps
System maintenance
setup: initialize and diagnose the environment
metrics: inspect token usage, runtime, and call statistics
Additional note:
workspace is a core autor capability, but it is not currently implemented as a standalone skill file; use autor ws ... or the related MCP tools when needed
- If the task involves external literature acquisition and PDF downloads, switch explicitly to
autodownload instead of treating it as ordinary local-library search
- For Records-service workflows, prefer REST endpoints over CLI. Treat
/download, /search, and /fetch as long-running acquisition phases rather than quick metadata calls
3. Other core features
Beyond the skills, explain the underlying system capabilities autor provides:
- PDF -> Markdown: MinerU preserves LaTeX formulas and structured text; image attachments are stripped and generated review figures must use
autor plot / autor/plot.py
- Layered loading: L1 metadata, L2 abstract, L3 paper-level conclusion card, L4 full text
- Auditable retrieval: node-level SQLite FTS5 over metadata and
paper.md, with bundle/trace/verify artifacts for provenance
- Explore's isolated data flow: build external exploration libraries from OpenAlex without affecting the main library
- Citation graph: references, citations, and shared-citation analysis
- Multi-source import: Endnote, Zotero, PDF, Markdown
- Workspace isolation: writing, notes, and drafts are all written under
workspace/
- Multi-agent compatibility: Claude Code, Copilot, Codex, OpenClaw, Cline, Cursor, Windsurf, and MCP clients
- Records service integration: the Records-backed AutoDownload service can be used as an external REST service for PubMed retrieval and PDF downloads
- Multiple document types: regular papers, theses, technical reports, lecture notes, and other document types
Recommended answer structure
When the user asks a broad question, organize the answer in this order by default:
- Explain in one sentence what autor is
- Give the three usage modes
- List the skills by group
- Add the features that differentiate autor from ordinary paper managers
- Recommend the next most relevant skill or CLI command if the user wants to continue
When to switch to another skill
- The user wants to install dependencies or troubleshoot the environment: switch to
setup
- The user wants to find papers: switch to
search
- The user wants to read a paper: switch to
show
- The user wants to ingest new literature: switch to
ingest
- The user wants topic analysis: use
search, research, or explore to build an evidence-grounded theme map
- The user wants to expand an existing workspace through external acquisition: switch to
autodownload
- The user wants to build a new workspace from a topic / outline and needs external literature acquisition: switch to
autodownload
- The user wants to revise structure, classify papers, and design tables before writing a review: switch to
plan
- The user wants to formally draft review text after
/plan: use paper-writing section by section, or follow the user's explicitly chosen drafting workflow
- The user wants a more open-ended review or direct section drafting: switch to
literature-review or paper-writing
- The user wants reviewer-driven manuscript revision with minimum necessary changes: switch to
update
- The user wants to verify citations: switch to
citation-check
Boundary facts
- autor is a research terminal / knowledge infrastructure, not just a PDF manager
- Skills are agent-oriented workflow wrappers; they do not exhaust the system's underlying capabilities
- User-facing outputs should be written to
workspace/ by default, not the project root or the autor/ package directory
- If the user's question is already specific enough, move quickly into the corresponding specialized skill instead of staying at the overview layer