Skip to main content
skill CX Codex

crossthink — Claude Code skill: what it does, how to install

crossthink — a skill for Claude Code that borrows operational mechanisms from distant fields when technical options get stuck on a single blocker.

MIT 121 installs ★ 1,470 updated 2026-10-25

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add LilMGenius/paperthin --skill crossthink -a claude-code
Open on GitHub

What crossthink does

The skill states the problem constraints and catalogs direct options before mapping mechanisms from five or more distant domains like logistics, biology, or finance. Each mapped candidate specifies who performs which step and defines a concrete failure condition (breaks-where).

Entries already present in the direct list or lacking a concrete causal operation get rejected with reasons. The output prioritizes mechanisms that remove the blocker, followed by those that isolate damage while it stands. The skill remains strictly advisory and changes no project code.

What it helps with

  • break out of single-option deadlocks caused by a blocked dependency
  • extract working operations from biology, law, or finance
  • identify specific failure points for cross-domain transfers
  • isolate operational damage while waiting on external approvals

Example prompts for the agent

  • All options currently depend on legal review. Run crossthink on this bottleneck.
  • Map mechanisms from distant domains to handle database failover without consensus.

How to install crossthink

1Skills CLI (Vercel)

npx skills add LilMGenius/paperthin --skill crossthink -a claude-code

2Manually: put the files in place

~/.claude/skills/crossthink/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/crossthink/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about crossthink

How to install crossthink in Claude Code?

Run npx skills add LilMGenius/paperthin --skill crossthink -a claude-code in the terminal. Alternatively, put the skill definition into ~/.claude/skills/crossthink/SKILL.md.

What does crossthink do?

It writes down the problem baseline, transfers working operations from distant disciplines, and outputs only the new options alongside their failure conditions.

How does crossthink differ from macrothink and factchk?

The tool transfers mechanisms horizontally across fields, while macrothink reframes scope vertically and factchk validates source claims.

Source text

SKILL.md · MIT
---
name: crossthink
description: "When every option on the table is a variation of one route or waits on the same sign-in, approval, person or dependency, list the options that frame already gives, then map working mechanisms from five or more distant domains onto the problem and report the ones the first list lacked, each with where it breaks. Use when work is stuck inside one frame, when the user asks how another field or industry would solve it, or asks for an analogy that actually transfers. Not for re-reading a decision at a larger scale (macrothink), listing options when no single frame is stuck, or checking claims (factchk)."
---

List what the problem's own frame already offers, then borrow working mechanisms from distant domains and keep only the ones that list lacked.

## Goal

A problem stuck in one frame keeps producing variations of one route. A mechanism that already works in another domain can open a route the frame hides, but only when the transfer is operational: the source mechanism is real, the target mapping says who or what does which step here, and the breaks-where names the condition in this setting under which it fails. The run returns the new mechanisms with their breaks-where, or says honestly that the mapping found none.

The mapping does not replace direct generation. On the problems it was tried on, direct generation found more usable mechanisms, and the mapping still found ones direct generation missed, most of them about keeping the work safe while the blocker stands (reconciliation, staging against a stand-in, escalation, findings kept as text).

## Workflow

1. **State the problem and its frame.** Write the goal, the hard constraints, the stated unknowns and the frame that narrows the options now: the route every current option takes and the step it waits on, if any. Keep the problem's facts; drop the session's preferred answer.
2. **List the frame's own options first.** Write the distinct mechanisms the problem already yields directly, one line each: how it works here and what it needs. This list is the baseline the mapping is judged against, so write it before choosing any domain.
3. **Map distant domains until five rows are accepted.** Pick domains far from the problem's field (logistics, biology, law, finance, film, games, hardware, cognitive science, real estate, or others). For each, fill three cells:
   - **source mechanism**: an operation or causal process that actually works in that domain, stated as what happens, not as a slogan;
   - **target mapping**: the concrete option it yields here, saying who or what performs which step and what it requires;
   - **breaks-where**: the specific condition in this setting under which the transfer fails, and how to handle it when it can be handled. "May not work" or a cost alone is not a condition.
4. **Reject what does not transfer.** A row with an empty cell, a mapping that only renames words with no operation behind it, and a source mechanism you cannot state as a working operation (reject it as unverified; this pass does not check sources) go to a rejected list with the reason.
5. **Fold and subtract.** Fold rows that share one working operation into one mechanism; a different domain, vendor, tool, parameter or wording does not make a new one. Move every mechanism already on the step 2 list to the rejected list with the reason `already direct: <its step 2 line>`. What remains is the result.
6. **Report.** Lead with the new mechanisms, each with its domain, target mapping and breaks-where: first those that remove the blocker, then those that limit its damage while it stands. Name the cheapest observation that would confirm or break the first one. Close with the rejected list. Zero new mechanisms is a valid result: say so and keep the step 2 list as the answer.

## Rules

- Direct options are written first and never replaced by the mapping; the mapping is judged only by what it adds.
- A rejected row is kept with its reason, never deleted.
- Never invent a fact about a source domain. Checking a claim against sources belongs to `factchk`.
- Re-reading the same decision at a larger scale, purpose or horizon is a vertical read and belongs to `macrothink`; this skill reads sideways, across domains.
- Read-only and advisory: it proposes options and changes nothing, and choosing among them stays with the main work.

## Verification

Before finishing, confirm:

1. The frame and the step 2 list were written before any domain was mapped.
2. At least five rows were accepted. Each has a source mechanism, a target mapping and a concrete breaks-where.
3. No reported mechanism is already on the step 2 list, and rows sharing one operation were folded.
4. Every rejected row, the already-direct ones included, is listed with its reason.
5. The report names one next observation, or states that the mapping added nothing.

© LilMGenius · MIT · LilMGenius/paperthin

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set48

The whole set →
whychk whychk — a skill for Claude Code and Codex that challenges freshly made choices. An isolated sub-session identifies unstated assumptions, demanding explanations one gap at a time. CX Codex +1 123installs re0-decide re0-decide — a skill for Claude Code that turns proposal materials into a traceable decision instrument without recommending outcomes. CX Codex +1 123installs re0-squash The re0-squash Claude Code skill condenses a GitHub repository into one signed commit, then recreates its remote. It runs on an explicit request to replace a repository’s history. CX Codex +1 123installs scopechk scopechk is a Claude Code skill that checks whether an item falls within a target repository scope. Run it before adding or moving a module into a repository. CX Codex +1 123installs re0-import re0-import — a Claude Code skill routes handed-over material into documents in an existing knowledge base. It follows that base’s own routing rules for notes, links, and corrections. CX Codex +1 123installs deid This Claude Code skill finds one person’s or organization’s data in a shared artifact and replaces matches with roles or placeholders. It runs before publication or when a repository needs de-identification. CX Codex +1 121installs paperthin Takes requests entered via /paperthin and forwards them directly to the re0-workflow router. CX Codex +1 120installs backprop backprop — a skill for Claude Code feeds lessons from errors and corrections back into source files. Runs after failed cycles, repeated user prompts, or during rule cleanup passes. CX Codex +1 120installs

Project sponsors

Companies that keep this analytics open