More screenshots
Works with the AI tools you already use
Claude Code
Cursor
Codex CLI
GitHub Copilot
Gemini CLI+20 moreClaude CodeCursorCodex CLIGitHub CopilotGemini CLI
OpenClaw
Manus
Grok+17 agentsPDF Batch Comparison & Outlier Finder
Compare one text-layer PDF baseline against a folder and surface repeated changes, one-off outliers and per-file diffs without uploading documents.
Free
PDF Batch Comparison & Outlier Finder
Example session with this skill installed
Compare baseline.pdf against the four text-layer PDFs in ./candidates. Save the report to ./qa-output. Do not modify the PDFs, and identify shared changes and frequency outliers.
- Read your context and instructions
- Compiled the pdf batch comparison
- Generated the data export
PDF Batch QA Report
- Status:
DIFFERENT - Baseline:
baseline.pdf - Candidates: 4
- Comparable: 4
- Review: 0
- Errors: 0
- Change groups: 3
This report compares extracted text layers. It does not prove visual or semantic equivalence. Common changes remain differences; outliers describe frequency, not correctness.
Candidate summary
| Candidate | Status | Changes | Outlier groups |
|---|---|---|---|
| document-001.pdf | DIFFERENT | 1 | 0 |
| document-002.pdf | DIFFERENT | 1 | 0 |
| document-003.pdf | DIFFERENT | 2 | 0 |
| document-004.pdf | DIFFERENT | 3 | 1 |
Aggregate change groups
COMMON_CHANGE — Fee
All 4 of 4 comparable files changed Fee: $10.00 to Fee: $12.00 on page 1. This remains a confirmed difference even though it appears throughout the batch.
SHARED_CHANGE — Region
2 of 4 files changed Region: East to Region: West on page 1. Neither variant is a strict-minority frequency outlier.
UNIQUE_CHANGE — Status
Only document-004.pdf changed Status: Draft to Status: Final. The file is flagged as the frequency outlier for this change group; the report does not claim which value is correct.
Artifacts written
result.json— authoritative structured resultbatch-summary.csv— sortable candidate statusesDIFF_REPORT.md— this readable reportfiles/*.diff.txt— one bounded text diff per candidateevidence-receipt.json— input and output identities
Next review steps
- Confirm whether the batch-wide fee change was intended.
- Check the two region variants against the source data.
- Review
document-004.pdfand its per-file diff before delivery.
pdf-batch-comparison-outlier-finder.csv
CSV · data export
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Reviewing a folder of similar PDFs one pair at a time makes it easy to miss a repeated template change or the single file that differs from the rest. This local tool compares one text-layer baseline against a bounded folder, groups shared changes, flags frequency outliers and creates a reproducible review packet. It does not perform OCR or visual pixel comparison.
Review a PDF batch without uploading it
Comparing one baseline against a folder one pair at a time makes repeated template changes and one-off outliers easy to miss. This local package turns text-layer PDFs from the same general template into a bounded, reproducible QA packet.
What one run creates
result.json— the authoritative machine-readable result.batch-summary.csv— a sortable file-status table.DIFF_REPORT.md— a readable batch summary.- One bounded text diff for every candidate.
evidence-receipt.json— input and artifact hashes.
What it distinguishes
The engine reports exact matches, confirmed text differences, review conditions and errors. Across reliable comparisons it groups common, shared, unique and mixed changes. A frequency outlier means different from the strict majority; it never means wrong.
Best for
- Automation freelancers checking generated client documents.
- Operations teams reviewing same-template statements, reports or certificates.
- Small QA teams that need a reproducible review packet before delivery.
Privacy and runtime
The package includes an independent CLI, a local stdio MCP adapter and a companion Skill. PDF parsing and comparison run locally in supervised subprocesses with time, memory and protocol limits. It does not upload documents, call web services, modify PDFs or require an account.
Requirements and limitations
Tested on Windows 11 with Python 3.11–3.14 and pypdf==6.18.1; the optional MCP adapter uses mcp==2.2.0. V1 compares extracted text layers only. It has no OCR or pixel-level visual comparison and does not prove semantic, table, legal or regulatory equivalence. Scanned, encrypted, ambiguous, resource-limited or truncated cases become REVIEW or ERROR, never MATCH.
How to install
Works the same in every agent - Claude, Cursor, Codex, Copilot and 20+ more.
- 1
Download the ZIP
Free skills download straight away. Paid skills unlock right after purchase.
- 2
Unzip into your skills folder
Every agent reads skills from one folder on your machine. Drop the unzipped folder in there.
- 3
Ask your agent to use it
Restart the agent if it was already running. It picks the skill up automatically - no config needed.
Skills folder by agent
Click the path to copy it. Create the folder if it does not exist yet.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
1 install
Downloaded by developers to date
Free forever
No account required to browse
Trust & safety
Security scanned
Verified clean 19 days ago
- Free to download with an account