Skill: report-analyzer

Analyzes large reports (PDF, PPTX) and produces a structured summary with key data and insights.

Target audience: product managers, marketers, analysts, finance professionals, C-level executives.

---

Workflow

Step 1 — Locate the input file

The user specifies the report file name in their message. Claude must locate this file.

Search order:

1. Check the current project working directory (

   input/

   folder or project root)
2. Check

   /mnt/user-data/uploads/

   (if the file was uploaded via the interface)
3. If the user provided a full path — use it directly

File search:

# Search by name in working directory and uploads
find /mnt/user-data/uploads/ -iname "*.pdf" -o -iname "*.pptx" 2>/dev/null
find . -iname "*.pdf" -o -iname "*.pptx" 2>/dev/null

If the file is not found — ask the user:

"Please specify the exact file name or path to the report. Supported formats: PDF, PPTX."

If the format is not supported:

"This skill works with PDF and PPTX files. Please provide a file in one of these formats."

Important: copy the file to

/home/claude/

before processing to avoid modifying the original.

---

Step 2 — Ask clarifying questions

Use

ask_user_input

for three questions simultaneously:

Question 1 (single_select): "What language should the output be in?"

• Russian
• English

Question 2 (single_select): "Analysis focus?"

• General overview
• Numbers and data
• Strategic takeaways
• Everything combined

Question 3 (single_select): "Output file format?"

• .md
• .pdf
• .docx

---

Step 3 — Read and process the file

For PDF:

import pdfplumber

with pdfplumber.open("path/to/file.pdf") as pdf:
    text = ""
    for page in pdf.pages:
        text += page.extract_text() or ""
        # Extract tables
        tables = page.extract_tables()

Also extract metadata:

from pypdf import PdfReader
reader = PdfReader("path/to/file.pdf")
meta = reader.metadata
pages_count = len(reader.pages)

For PPTX:

python -m markitdown presentation.pptx

If text extraction fails (scanned PDF) — use OCR:

import pytesseract
from pdf2image import convert_from_path
images = convert_from_path('file.pdf')
for image in images:
    text += pytesseract.image_to_string(image)

---

Step 4 — Perform analysis

Based on the extracted text, generate content according to the structure below.

Analysis rules:

• Extract only facts and data from the report — do not infer or fabricate
• All numbers must be exact — as in the original
• Formulate insights in the third person ("the authors note", "the report indicates")
• If data is contradictory — flag it explicitly
• Output document must not exceed 1.5 pages

How to identify insights (selection criteria):

An insight is a statement from the report that meets at least one criterion:

1. Non-obvious — the finding contradicts conventional wisdom or expectations ("contrary to popular belief...", "unexpectedly...")
2. Quantitative shift — a sharp change in a metric (growth/decline >20%, trend reversal, record)
3. Causal relationship — the report explicitly links cause and effect ("X led to Y", "the main driver is Z")
4. Forecast or warning — authors make a prediction or flag a risk with specific parameters
5. Actionability — a finding that enables a management decision (not just a description of the situation)

How to formulate an insight:

• One sentence, two at most
• Lead with the substance, not the context
• Include specific numbers where available
• Bad: "The AI market is growing" → Good: "The GenAI market grew 68% YoY to $127B, outpacing analyst forecasts by 15 p.p."

Number of insights: 5–7. If the report contains fewer significant findings — do not pad to 5 artificially.

Adaptation by focus:

• General overview — balanced across all sections
• Numbers and data — expanded data table, condensed insights
• Strategic takeaways — expanded insights, condensed data table
• Everything combined — all sections in full (within the 1.5-page limit)

---

Step 5 — Generate the output file

Use the structure from the "Output document template" section below.

By format:

• .md

  — write directly via

  create_file

• .pdf

  — first generate

  .md

  , then use the

  pdf

  skill (read

  /mnt/skills/public/pdf/SKILL.md

  )

• .docx

  — first generate

  .md

  , then use the

  docx

  skill (read

  /mnt/skills/public/docx/SKILL.md

  )

File naming:

report-summary_REPORT-NAME_YYYY-MM-DD.extension

Example:

report-summary_mckinsey-ai-trends_2026-03-18.md

Save the file to

/mnt/user-data/outputs/

and deliver to the user via

present_files

.

---

Output document template

# Report Summary: [Report Title]

**Analysis date:** [date]

---

## Report metadata

| Parameter | Value |
|-----------|-------|
| Title | [full title] |
| Author / source | [company or author] |
| Publication date | [date or year] |
| Length | [page count] |
| Original language | [language] |
| Type | [consulting / research / analytics / market review / other] |

---

## Executive Summary

[3–5 sentences: core thesis of the report, main argument, key conclusion]

---

## Key figures and data

| Metric / indicator | Value | Context |
|--------------------|-------|---------|
| [metric 1] | [value] | [explanation] |
| [metric 2] | [value] | [explanation] |
| ... | ... | ... |

---

## Key insights

1. [insight 1]
2. [insight 2]
3. [insight 3]
4. [insight 4]
5. [insight 5]

---

## Report structure

| Section | Summary |
|---------|---------|
| [Section 1] | [1–2 sentences] |
| [Section 2] | [1–2 sentences] |
| ... | ... |

---

What to avoid

• Do not add personal assessments or recommendations — only report content
• Do not exceed 1.5 pages in the output document
• Do not use filler phrases ("Sure, here's the analysis...")
• Do not retell the entire report — highlight what matters
• Do not round or alter numbers from the original
• Do not create the file until all 3 questions are answered

---

Dependencies

pip install pdfplumber pypdf markitdown --break-system-packages

For PPTX:

pip install "markitdown[pptx]" --break-system-packages

For OCR (scanned PDFs):

pip install pytesseract pdf2image --break-system-packages