pdf-text-extractor-readability-classification
Sub-skill of pdf-text-extractor: Readability Classification.
Best use case
pdf-text-extractor-readability-classification is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Sub-skill of pdf-text-extractor: Readability Classification.
Teams using pdf-text-extractor-readability-classification should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/readability-classification/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How pdf-text-extractor-readability-classification Compares
| Feature / Agent | pdf-text-extractor-readability-classification | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Sub-skill of pdf-text-extractor: Readability Classification.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
SKILL.md Source
# Readability Classification
## Readability Classification
Before extracting text from a large PDF collection, classify each PDF's readability
using `enrich-readability.py`. This determines which extraction strategy to use:
| Classification | Meaning | Extraction strategy |
|---------------|---------|-------------------|
| `machine` | Text layer present, directly extractable | pdfplumber / PyMuPDF |
| `ocr-needed` | Scanned image, no text layer | tesseract / doctr / azure-doc-intelligence |
| `mixed` | Some pages machine-readable, some scanned | Hybrid — extract text pages, OCR image pages |
| `error` | Corrupted or unreadable | Skip; log for manual review |
**Key finding**: 27-30% of project PDFs are scanned with no text layer. Attempting
direct text extraction on these returns empty strings — always classify first.
### Final Corpus State (WRK-1277, 2026-03-17)
| Classification | Count | Percentage |
|---------------|-------|-----------|
| native | 623,455 | 60.3% |
| machine | 278,899 | 27.0% |
| ocr-needed | 92,042 | 8.9% |
| missing | 27,476 | 2.7% |
| error | 6,221 | 0.6% |
| mixed | 5,246 | 0.5% |
| **Total classified** | **1,033,933** | **96.7%** |
Error reduction: 296,626 → 6,221 (97.9% recovery). Remaining errors are genuine
edge cases (corrupt PDFs, missing files, extremely complex documents).
### Classification Method
**Use pdftotext (poppler) for batch classification** — not pdfplumber:
```bash
# Classify all PDFs with parallel workers (resume-safe)
uv run --no-project python scripts/data/document-index/enrich-readability.py \
--workers 10 --resume
```
Use `--workers 10` for bulk enrichment to parallelize across CPU cores. The `--resume`
flag skips already-classified entries, making it safe to restart after interruption.
> **WARNING (WRK-1277)**: The original `enrich-readability.py` used pdfplumber in
> `ProcessPoolExecutor` — this hung in D-state on NTFS/NFS mounts. The proven pattern
> is pdftotext via `subprocess.run(timeout=30)` with 8 workers (see
> `pdf/pdftotext-poppler` sub-skill for code). Throughput: ~49 files/sec vs ~1.3 with
> pdfplumber.Related Skills
hermes-kanban-readability
Reapply the Hermes Kanban dashboard readability customizations (clickable bare URLs in card descriptions + readable card-text font, a visible horizontal scrollbar so all columns are reachable) as a user-override plugin that survives hermes-agent updates. Use when the Kanban board reverts to the Mondwest display font / plain-text Source URLs after a hermes update, or when bootstrapping a machine whose ~/.hermes was wiped.
sodir-data-extractor
Extract and process Norwegian Petroleum Directorate field and production data from SODIR
bsee-data-extractor
Extract and process BSEE (Bureau of Safety and Environmental Enforcement) data including production, WAR (Well Activity Reports), and APD (Application for Permit to Drill) data. Use for querying production data, well activities, drilling permits, completions, and workovers by API number, block, lease, or field with automatic data normalization and caching.
context-compaction-handoff
Guardrails for resuming work after context compaction or transcript handoff blocks; prioritize the latest real user request over stale summarized tasks and verify before answering.
boundary-policy-classification-by-role
Classify artifacts as durable vs transient by their functional role rather than directory path, using multi-layer architectural validation
provider-audit-bootstrap-and-path-classification
Fix provider-session ecosystem audit failures caused by source-checkout imports and over-aggressive symbolic-path classification.
wiki-context
Auto-query llm-wiki domains for relevant context before executing domain tasks
pdf-text-extractor
Extract text from PDF files with intelligent chunking and metadata preservation. For batch extraction (1K+ PDFs), use pdftotext (poppler) via subprocess — see pdf skill Tool Selection table. For single-doc quality, use Codex or PyMuPDF. Supports technical documents, standards libraries, research papers, or any PDF collection.
pdf-pdftotext-poppler
Sub-skill of pdf: pdftotext (Poppler) (+2).
data-context-extractor
Generate or improve a company-specific data analysis skill by extracting tribal knowledge from analysts
pretext
Use when building creative browser demos with @chenglou/pretext — DOM-free text layout for ASCII art, typographic flow around obstacles, text-as-geometry games, kinetic typography, and text-powered generative art. Produces single-file HTML demos by default.
skill-chain-context-optimization
Refactor large or frequently-run skills into context-efficient chains using isolated execution, file-backed handoffs, minimal summaries, and runtime-aware command substitution.