How do I install the Cross Doc Check skill?

Run npx skillmds@latest add alawein/cross-doc-check in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Cross Doc Check safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Cross Doc Check?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Cross Doc Check free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Cross Doc Check?

alawein (@alawein) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Cross Doc Check by alawein — Agent Skill

Name: Cross Doc Check
Author: alawein

Cross Doc Check

Use when two docs are each supposed to state the same policy or rule, to check whether they actually agree. Flags a real contradiction or an unexplained scope gap, never a wording difference where the substance already agrees. Distinct from doc_truth, which checks that a named reference resolves, not whether two docs agree with each other.

alawein 353beb7 2.4 KB Updated Sep 17, 2026

File contents

cross-doc-check

doc_truth checks that a prompt or tool named in prose still exists. Nothing checks whether two docs that are each supposed to state the same rule actually agree. This repo has already shipped that exact bug once: a cited number in one doc quietly disagreed with the same number stated elsewhere, caught by luck during a wider review pass, not by anything built to find it.

When to use it

Given two or more docs that are each supposed to state the same or overlapping policy, before trusting they agree.
Right after editing one of a pair of docs that share a rule, before the edit ships.
Skip it when the docs cover different, non-overlapping topics; there is nothing to reconcile.

Required inputs

The specific docs to compare, named by the caller. This prompt does not guess which docs are supposed to agree.

Steps

Read every named doc in full.
From each, extract the concrete, checkable rules it states: a number, a required or optional call, a threshold, a named exception. Set aside scene-setting prose that states no rule.
Compare the extracted rules across docs. Report only two kinds of finding: a direct contradiction, where both docs cover the same case and require a different action, or an unexplained scope gap, where one doc's rule has no counterpart in the other and neither doc says why. A rule stated in different words, with the same substance, is not a finding.
For each finding, name the exact location in each doc and the conflicting values or rules.

Output format

Findings: one line each, the two locations and what conflicts. Empty if none.
Verdict: agree (no findings) or disagree, with the count.
Read-only: it reports the mismatch; it does not edit either doc.

Stop conditions

Every rule extracted from every named doc has been checked against every other named doc.
A wording difference alone, with the same substance, is never reported as a finding.
Stop once every extracted rule is accounted for. Do not judge whether a rule is itself a good one; that is out of scope.

---name: cross-doc-checkdescription: Use when two docs are each supposed to state the same policy or rule, to check whether they actually agree. Flags a real contradiction or an unexplained scope gap, never a wording difference where the substance already agrees. Distinct from doc_truth, which checks that a named reference resolves, not whether two docs agree with each other.---# cross-doc-check`doc_truth` checks that a prompt or tool named in prose still exists. Nothing checks whether twodocs that are each supposed to state the same rule actually agree. This repo has already shippedthat exact bug once: a cited number in one doc quietly disagreed with the same number statedelsewhere, caught by luck during a wider review pass, not by anything built to find it.## When to use it- Given two or more docs that are each supposed to state the same or overlapping policy, before trusting they agree.- Right after editing one of a pair of docs that share a rule, before the edit ships.- Skip it when the docs cover different, non-overlapping topics; there is nothing to reconcile.## Required inputs- The specific docs to compare, named by the caller. This prompt does not guess which docs are supposed to agree.## Steps1. Read every named doc in full.2. From each, extract the concrete, checkable rules it states: a number, a required or optional call, a threshold, a named exception. Set aside scene-setting prose that states no rule.3. Compare the extracted rules across docs. Report only two kinds of finding: a direct contradiction, where both docs cover the same case and require a different action, or an unexplained scope gap, where one doc's rule has no counterpart in the other and neither doc says why. A rule stated in different words, with the same substance, is not a finding.4. For each finding, name the exact location in each doc and the conflicting values or rules.## Output format- Findings: one line each, the two locations and what conflicts. Empty if none.- Verdict: agree (no findings) or disagree, with the count.- Read-only: it reports the mismatch; it does not edit either doc.## Stop conditions- Every rule extracted from every named doc has been checked against every other named doc.- A wording difference alone, with the same substance, is never reported as a finding.- Stop once every extracted rule is accounted for. Do not judge whether a rule is itself a good one; that is out of scope.

Frequently asked questions