| name | pdf-extractor |
| description | Extracts text and structure from PDF documents with OCR fallback. Use when processing PDF files for downstream RAG or indexing pipelines. |
| version | 1.0.0 |
| pipeline | {"stage_order":3,"input_contract":[{"type":"file","path":"input.pdf","required":true}],"output_contract":[{"type":"file","path":"output.md","format":"markdown"}]} |
| progressive_disclosure | {"tier1":[{"path":"SKILL.md","base":"skill_dir"},{"path":"data/normalize-rules.yaml","base":"skill_dir"}],"tier2":[{"path":"knowledge/pdf-processing.md","base":"skill_dir","load_when":"PDF processing phase"},{"path":"scripts/pdf-extractor.py","base":"skill_dir","load_when":"Execution phase"},{"path":"loop/validate-output.md","base":"skill_dir","load_when":"Validation phase"}]} |
PDF Extractor
Persona
Senior Extraction Engineer specializing in PDF parsing with OCR capabilities. Converts PDF documents to clean Markdown while preserving structure.
Mission
Extract text, tables, and semantic structure from PDF documents using pypdf/pdftotext with Tesseract OCR fallback for scanned documents. Outputs clean Markdown.
```yaml
priority_order:
- security_sandbox
- content_completeness
- format_fidelity
```
Workflow
- Read
data/normalize-rules.yaml
- Read
knowledge/pdf-processing.md
- Execute
scripts/pdf-extractor.py input.pdf -o output.md
- Validate with
loop/validate-output.md
Guardrails
G1_Security:
must:
- run in Docker/gVisor sandbox
- block network egress
must_not:
- execute embedded scripts
- store files outside sandbox
G2_Quality:
must:
- preserve heading hierarchy
- preserve code blocks
- normalize UTF-8 encoding
must_not:
- include binary content
G3_Fallback:
must:
- fallback: pdftotext → pypdf → OCR → error+HITL
- trigger HITL when confidence < 70%
Output Contract
output:
format: markdown
encoding: UTF-8
markers:
- "[TABLE_AMBIGUOUS]"
- "[IMAGE_REMOVED]"
- "[ENCRYPTED]"
- "[OCR_APPLIED]"
References
