ocr-quality-assurance — Claude Code subagent: how to install
Inspects OCR-processed text against original scan images to catch omitted data and bad markdown syntax. Useful as the last validation gate in document processing.
npx claude-code-templates@latest --agent ocr-extraction-team/ocr-quality-assuranceWhat ocr-quality-assurance does
This subagent acts as the fifth step in a document digitisation pipeline. It follows visual inspection, textual diffing, contextual grammar passes, and markdown structuring. Its sole duty is auditing the output against the source picture so no numbers or punctuation get lost.
The subagent uses Read and Write tools to inspect headers, lists, links, and markdown table alignment. If an ambiguous symbol appears, it will not guess. Instead, it marks the fragment with [REVIEW NEEDED: description] to call in a human reviewer.
Validation concludes with a structured summary marked as APPROVED, APPROVED WITH NOTES, or REQUIRES HUMAN REVIEW. Without the original image file available, the subagent cannot verify transcription fidelity.
What it helps with
- audit OCR text output against source image files
- verify markdown tables, lists, and headings for syntax errors
- detect dropped punctuation, numerals, and formatting emphasis
- tag ambiguous text segments for manual human inspection
Example prompts for the agent
- Validate
report.mdagainstscan.pngusing ocr-quality-assurance - Audit the transcribed tables in
table.mdagainstpage1.jpgand produce a QA report - Check
extracted_text.mdagainstoriginal.pngand flag any formatting discrepancies
How to install ocr-quality-assurance
1claude-code-templates
npx claude-code-templates@latest --agent ocr-extraction-team/ocr-quality-assurance2Manually: put the files in place
.claude/agents/ocr-quality-assurance.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/ocr-quality-assurance.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/ocr-quality-assurance.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about ocr-quality-assurance
How do I install ocr-quality-assurance in Claude Code?
Run npx claude-code-templates@latest --agent ocr-extraction-team/ocr-quality-assurance. This saves the configuration to .claude/agents/ocr-quality-assurance.md.
Where does this subagent fit in the OCR pipeline?
It operates as the fifth and final stage, taking over after visual analysis, text comparison, grammar checking, and markdown formatting are complete.
Which outcome statuses can the validation report return?
The final report issues APPROVED, APPROVED WITH NOTES, or REQUIRES HUMAN REVIEW. Uncertain passages are marked with [REVIEW NEEDED: description].
