agentleFS
ch4570ch4570/vulpora/skills/pdf-qa/SKILL.md

pdf-qa

Skill1 starsChanged 56 days agoApache-2.0View on GitHub

Render every page of an existing PDF to PNG and gate delivery on recorded visual inspection. Use after generating or changing a PDF, before claiming a PDF report, document, diagram, or slide deck is complete, or when diagnosing clipping, overlap, broken glyphs, pagination, page-balance, or table-spacing defects. This skill validates PDFs; it does not author or repair their source.

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run scripts/preflight.py to verify that Poppler provides pdfinfo and at least one of pdftocairo or pdftoppm.

    line 15
  2. Very specific:

    Run scripts/render-pages.py --input <file.pdf> --output-dir <qa-dir> --source-ref <editable-source>.

    line 16
  3. Very specific:

    Create one pdf-qa.inspection/v1 JSON record covering every rendered page and every required check.

    line 18
  4. Very specific:

    If the final status is revise, return the issues to the source-producing skill, regenerate the PDF, and repeat from step 2.

    line 20
  5. Very specific:

    Treat the initial render manifest as revise.

    line 24
  6. Distinctive:

    Keep the PDF, editable source reference, rendered pages, render manifest, inspection record, and final QA manifest linked and available for …

    line 29

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

PDF QA

Act as a gatekeeper. Never certify visual quality from PDF metadata, text extraction, OCR, or successful rendering alone.

Enforce the visual loop: Generate -> Render -> Inspect -> Fix -> Re-render.

Workflow

  1. Preserve the editable source and identify its relationship to the PDF. If the source is unavailable, record that limitation instead of inventing one.
  2. Run scripts/preflight.py to verify that Poppler provides pdfinfo and at least one of pdftocairo or pdftoppm.
  3. Run scripts/render-pages.py --input <file.pdf> --output-dir <qa-dir> --source-ref <editable-source>; use 180 DPI unless the artifact needs closer inspection.
  4. Inspect every emitted page PNG visually. Use view_image when available. Read references/qa-checklist.md for the checks and inspection record shape.
  5. Create one pdf-qa.inspection/v1 JSON record covering every rendered page and every required check. Record the rendered image SHA-256 so stale evidence cannot pass.
  6. Run scripts/record-inspection.py --manifest <qa-manifest.json> --inspection <inspection.json> --output <qa-final.json>.
  7. If the final status is revise, return the issues to the source-producing skill, regenerate the PDF, and repeat from step 2. Never patch the rendered PNG as the fix.

Gate Rules

  • Treat the initial render manifest as revise; rendering is evidence preparation, not inspection.
  • Require visual evidence for clipping, overlap, glyph_integrity, page_balance, and table_spacing on every page.
  • Accept only visual_image_review, vision_model_review, or human_visual_review as inspection methods. Do not treat text-only or OCR-only methods as visual evidence.
  • Set pass only when every required check on every rendered page is explicitly recorded as pass against the current image digest.
  • Set revise when any check fails. Include a concrete page-level detail for every result.
  • Keep the PDF, editable source reference, rendered pages, render manifest, inspection record, and final QA manifest linked and available for reproduction.

Completion

Report PDF QA complete only when qa-final.json has status: pass, all pages were rendered at the recorded DPI, all image digests match the inspection, and no required check is missing. Otherwise report revise and the issue list.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.