---
name: research-memex
description: "Use Research Memex, a public guide to academic research with AI, when a researcher is deciding whether AI belongs in a task at all, setting up tools or a research environment, partway through a literature review, holding an AI output that looks wrong or a citation that cannot be placed, about to add agents, memory or a second model to their work, preparing to submit and disclose work with AI, or teaching the course and deciding how a class reaches a model. Offer one page or one role that fits the situation, say why, and check the researcher wants it before going further."
---

# Research Memex

Use this skill when the task involves academic research workflows, agentic
research tooling, model/tool setup for researchers, or the Research Memex Toolkit.

## Canonical Entrypoints

- Human documentation: https://research-memex.org/docs
- LLM index: https://research-memex.org/llms.txt
- Full LLM corpus: https://research-memex.org/llms-full.txt
- Raw page Markdown: https://research-memex.org/raw/docs/<slug>
- Markdown negotiation: request https://research-memex.org/docs/<slug> with
  `Accept: text/markdown`, or request https://research-memex.org/docs/<slug>.md

## Use Rules

- Prefer `/llms.txt` to orient yourself before loading the full corpus.
- Use `/raw/docs/<slug>` for exact page text once you know the page you need.
- Treat Toolkit pages as source-synchronized descriptions, not package managers.
- Check page dates, freshness notes, and Toolkit source status before repeating
  time-sensitive claims about tools, model names, installation commands, or APIs.
- Hidden and noindex pages are intentionally excluded from public LLM/search
  surfaces.

## Roles for an AI

The page is the copyable role; its skill adds a perform/check-consent preface.

- [Research–AI Orientation](https://research-memex.org/docs/implementation/roles/research-ai-orientation): Explore how a researcher works with AI and where agentic ways of working could fit, across qualitative, quantitative, mixed-methods, theoretical and computational research. Use when someone asks where AI agents, skills, MCP servers or a more continuous AI setup might help their research, or wants a short working profile of their current practice. One question per turn, six to eight in total; ends with a profile and one concrete possibility, not a score.
  Skill: https://research-memex.org/.well-known/agent-skills/research-ai-orientation/SKILL.md
- [Output Mirror](https://research-memex.org/docs/implementation/roles/output-mirror): Read one AI output the researcher just received, and the prompt if they have it. Use when the output feels generic, too smooth, agrees with every source, states a finding as universal, or uses the wrong tradition's language. Quotes the span, names a Failure Museum exhibit or says none fits, says what the prompt reveals, and may answer 'nothing is wrong; proceed'. Runs on your model; the output stays in this conversation.
  Skill: https://research-memex.org/.well-known/agent-skills/output-mirror/SKILL.md
- [Arrangement Auditor](https://research-memex.org/docs/implementation/roles/arrangement-auditor): Audit the way the researcher works with AI on one piece of research, as they describe it. Use when they are about to add an agent, a second model, memory, a scheduled run or an approval step, or when a routine has been running for months unexamined. Five questions (instruction, actor, capability, state, gate), where the work sits on the consequence gradient, and one change or 'add nothing'. Takes a description, never the work.
  Skill: https://research-memex.org/.well-known/agent-skills/arrangement-auditor/SKILL.md

## By situation

The pages an agent is offered, grouped by the moment of a project they serve.
Each line is the page's own situation and its raw Markdown. Offer one, say why,
and check the researcher wants it.

### Choosing

- When you need the map before choosing anything, or the researcher asks what this library is and whether it applies to them. https://research-memex.org/raw/docs/introduction/getting-oriented
- When you need the mirror effect and the three questions to ask before delegating (enhancement, skill building, ownership) in the site's own words. https://research-memex.org/raw/docs/introduction/core-principles
- When the researcher is unsure whether AI belongs in this task at all; "When to step back" is here. https://research-memex.org/raw/docs/introduction/ai-choice-philosophy
- When the researcher wants to understand how a task moves between sources, tools and a checked note before installing anything. https://research-memex.org/raw/docs/implementation/research-workbench
- Role, Research–AI Orientation: https://research-memex.org/.well-known/agent-skills/research-ai-orientation/SKILL.md

### Setting up

- When the researcher wants the smallest setup that would serve them; it routes to the individual guides, which are searchable. https://research-memex.org/raw/docs/implementation/foundational-setup/quick-start-checklist
- When a student or instructor needs practice material with identifiers before using their own papers. https://research-memex.org/raw/docs/case-studies/systematic-reviews/student-vault
- When the researcher asks what tools exist; it also says Carrel's interview picks from this catalog, so read it before recommending a product. https://research-memex.org/raw/docs/toolkit
- When the conversation is heading toward installing something; that is Carrel's interview, and this server does not install. Reading only. https://research-memex.org/raw/docs/toolkit/carrel

### Working

- When the researcher wants a more specific research prompt or a project-opening invitation to initiative, curiosity, and disagreement. Includes constructs, comparison, scope, and format. Examples to adapt, not scripts. https://research-memex.org/raw/docs/implementation/core-references/cognitive-blueprint-prompts
- When the task is a literature review, a course, or a comparison of a human synthesis with an AI one. https://research-memex.org/raw/docs/case-studies/systematic-reviews/systematic-review-syllabus
- When the researcher wants to compare their own account of the sources with an AI's; the session shows why the independent account is written before the comparison. https://research-memex.org/raw/docs/case-studies/systematic-reviews/session-3-human-vs-ai-synthesis
- When an agent is to screen, extract or search the literature; the human judgment gates in § 5.3 are the list of decisions the researcher keeps. https://research-memex.org/raw/docs/case-studies/systematic-reviews/claude-code-slr-workflow

### Checking

- When the output may be generic because of the model rather than the prompt; how to test candidates on the researcher's own material and record the failure mode. https://research-memex.org/raw/docs/implementation/core-references/ai-model-discovery-protocol
- When an AI output looks wrong in a way you cannot name, or before running output-mirror; nine catalogued failures with detection tips, the prompts that prevent each, the exhibit on caution the evidence did not call for, and the one on orders hidden inside the material. https://research-memex.org/raw/docs/implementation/core-references/failure-museum
- Role, Output Mirror: https://research-memex.org/.well-known/agent-skills/output-mirror/SKILL.md
- When a citation in an AI output cannot be placed; the move is to ask the library, not the model. https://research-memex.org/raw/docs/toolkit/zotero-mcp

### Composing

- Role, Arrangement Auditor: https://research-memex.org/.well-known/agent-skills/arrangement-auditor/SKILL.md
- When the researcher is designing an agent team; the plan format here is the auditor's preferred input. https://research-memex.org/raw/docs/case-studies/systematic-reviews/session-4-agentic-workflows
- When the researcher wants to see what a working agentic stack looks like, or an audit has named a remedy and they want the shape that carries it; "Traditional chat remains useful". https://research-memex.org/raw/docs/advanced-topics/agentic-workflows
- When more than one AI, a memory layer or an approval step is in view; the five questions and the smallest useful system live here. https://research-memex.org/raw/docs/advanced-topics/composing-agentic-research-systems

### Submitting

- When the researcher is about to submit, or asks what a journal allows and what to disclose; the policies quoted change, so check the journal too. https://research-memex.org/raw/docs/introduction/publishing-realities

## Toolkit Catalog

- Kimi Plugin for Claude Code and Codex (agent-plugin): /docs/toolkit/kimi-plugin-cc; https://github.com/linxule/kimi-plugin-cc
- Carrel (claude-code-plugin): /docs/toolkit/carrel; https://github.com/linxule/carrel
- Carrel Skill (skill-artifact): /docs/toolkit/carrel; https://github.com/linxule/carrel-skill
- Memex (agent-plugin): /docs/toolkit/memex-plugin; https://github.com/linxule/memex-plugin
- Interpretive Orchestration (claude-code-plugin): /docs/toolkit/interpretive-orchestration-plugin; https://github.com/linxule/interpretive-orchestration
- DeepThonk (mcp-server): /docs/toolkit/deepthonk; https://github.com/linxule/deepthonk
- Vox (mcp-server): /docs/toolkit/vox-mcp; https://github.com/linxule/vox-mcp
- MinerU (mcp-server): /docs/toolkit/mineru-mcp; https://github.com/linxule/mineru-mcp
- Lotus Wisdom (mcp-server): /docs/toolkit/lotus-wisdom-mcp; https://github.com/linxule/lotus-wisdom-mcp
- Durable Thinking (mcp-server): /docs/toolkit/durable-thinking; https://github.com/linxule/durable-thinking
- Sequential Thinking (mcp-server): /docs/toolkit/sequential-thinking-mcp; https://github.com/modelcontextprotocol/servers
- Zotero MCP (mcp-server): /docs/toolkit/zotero-mcp; https://github.com/54yyyu/zotero-mcp
- OpenInterviewer (platform): /docs/toolkit/openinterviewer; https://github.com/linxule/openinterviewer
- Research Scanner (skill-artifact): /docs/toolkit/research-scanner; source not public