Inputs
Provide these when prompted. The skill asks for anything missing before it runs.
Required
- A list of company or profile records
Optional
- Match keys to prioritize (domain, email)
- A confidence threshold
- Fields that must match exactly
Agent Skill
Duplicate Record Review
Identify likely duplicate profiles or companies using available record evidence.
Version 1.0 Updated June 2026 SKILL.md MIT 6 min read
Overview
What this skill does
Duplicates creep into every list from multiple sources and re-scrapes, inflating counts and causing double outreach.
This skill groups likely duplicates using the evidence available, reports match confidence and which fields agree or conflict, and recommends a merge or manual review — describing matches as likely rather than certain unless they are exact.
When to use it
Best used for
- De-duplicating a merged or re-scraped list
- Catching the same company under different names
- Preventing double outreach to one contact
- Producing a reviewable merge plan
Know the limits
When not to use this skill
- The records share no comparable identifiers
- You need an irreversible automatic merge (always review)
- The list is already known to be unique
Outputs
One record per duplicate group with a consistent, inspectable schema.
duplicate_groupRecords believed to refer to the same entity.
match_confidenceLikelihood the group is a true duplicate.
matching_fieldsFields that agree across the group.
conflicting_fieldsFields that disagree.
merge_recommendationSuggested surviving record and merges.
manual_review_statusWhether a human must decide.
Example
Example
A likely duplicate pair.
Input
A: Acme | acme.com | Maria Chen | maria@acme.com
B: ACME Inc | acme.com | M. Chen | maria@acme.comOutput
duplicate_group: A + B
match_confidence: high (exact email + domain)
matching_fields: email; domain
conflicting_fields: name format (Maria Chen vs M. Chen)
merge_recommendation: Keep A; merge name variant from B
manual_review_status: optional (high confidence)An exact email and domain make this a high-confidence duplicate; the only conflict is a name format, so the merge is safe with light review.
Setup
How to use the skill
General steps first, then notes for specific clients where verified.
- 1Download the file using the button below, or copy the Markdown.
- 2Place it in a directory named after the skill (e.g. skill-name/).
- 3Make sure the filename stays exactly SKILL.md.
- 4Add any references or assets included with the package.
- 5Load the skill into a compatible agent and provide the required inputs.
Claude Code
- 1Create a folder for the skill and save SKILL.md inside it.
- 2Place the folder where your project's skills are discovered.
- 3Reference the skill when you want it applied to your data.
Other compatible clients
- 1Confirm the client supports the open Agent Skills format.
- 2Load the SKILL.md file as instructed by that client.
- 3If skills are not auto-loaded, paste the Markdown as instructions.
Source
Full SKILL.md source
Read the rendered skill or copy the complete Markdown. The download is generated from this exact source.
Version 1.0 SKILL.md ~2 KB MIT
Duplicate Record Review
Purpose
Identify likely duplicate profiles or companies using available record evidence.
When to use this skill
- De-duplicating a merged or re-scraped list
- Catching the same company under different names
- Preventing double outreach to one contact
- Producing a reviewable merge plan
When not to use this skill
- The records share no comparable identifiers
- You need an irreversible automatic merge (always review)
- The list is already known to be unique
Required inputs
- A list of company or profile records
Optional inputs
- Match keys to prioritize (domain, email)
- A confidence threshold
- Fields that must match exactly
Rules
- Describe matches as likely unless an exact deterministic key matches.
- Report match confidence and the evidence behind it.
- Never auto-merge; recommend and flag for review.
- List conflicting fields explicitly.
- Prefer the most complete record as the survivor.
Process
- Compare records on available keys.
- Group likely duplicates.
- Assess confidence and list matching/conflicting fields.
- Recommend a survivor and merges.
- Set manual-review status.
Output format
Return one record per duplicate group with the following fields:
- duplicate_group
- match_confidence
- matching_fields
- conflicting_fields
- merge_recommendation
- manualreviewstatus
Validation
- Confirm only exact key matches are called certain.
- Confirm conflicts are listed for each group.
- Confirm no merge is performed automatically.
Limitations
- Without unique keys, matches are probabilistic.
- Distinct entities can share names; review before merging.
Before you rely on it
Safety and limitations
- Without unique keys, matches are probabilistic.
- Distinct entities can share names; review before merging.
- Review the output before acting on it.
- Do not upload confidential datasets to an external model without authorization.
- Outputs depend on the model and the source data and are not guaranteed to be accurate.
History
Changelog
- v1.0June 2026
- Initial release.
Questions
Agent Skill FAQ
How does it decide two records are duplicates?
It compares the available evidence — names, domains, emails, and other fields — and groups likely matches with a reason and merge guidance, leaving the final merge to you.
Will it merge records for me?
No. It recommends merges and flags review status; the actual merge stays a human decision to avoid data loss.
Do I need ProfileSpider to use this skill?
No. The skill works on any compatible data. ProfileSpider is one convenient way to produce that structured input.
Does running this skill send data to ProfileSpider?
No. Downloading or copying the file does not send any data to ProfileSpider. What happens afterward depends on the AI service you load it into.
Are Agent Skills the same as prompts?
No. A skill is a structured, reusable package — task, inputs, rules, process, and output format — so the workflow runs consistently and can be shared, versioned, and edited.