BB vlm-ocr
OCRs scanned or image-only corpora with vision-language models in three phases. Evaluate compares candidate systems against stratified human ground truth and selects one by measured CER or WER. Run builds the production pipeline, including model selection, image handling, prompts, architecture, batching, accuracy evaluation, and reproducibility. Clean corrects raw OCR with LLM and rule-based passes, quality diagnostics, multilingual handling, and span-level provenance. Use when choosing an OCR or VLM for a corpus, measuring accuracy on pages, transcribing scans at scale, or correcting raw OCR with QA or provenance logs. Born-digital documents with text layers go to doc-to-markdown.
OCRs scanned or image-only corpora with vision-language models in three phases.
As a process B 67/100 · Nearly there — weak spots: inputs and preconditions, execution cost
How to improve
- The SKILL.md body is over 5,000 tokens: move reference detail into references/ and load it when needed.
- Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
- A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.
Guard findings · 0
✓ No critical or high findings
Files scanned: 3. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- warning
body-longSKILL.md body ≈ 12200 tokens (recommended < 5000); move details to references/
Process rating: all ten parameters 67/100
- 0Inputs and preconditions. Does not say what the process needs to start
- 40Execution cost. Instruction body is 12200 tokens: crowds the task out of the window
- 50When it triggers. No condition that starts the skill
- 50Failures and branches. 0 branches, has a failure section
- 60Result and completion. Output format stated, no completion criterion
- 85Steps. 166 steps, 1 vague phrases
- 100Tools and files. No external tools needed
- 100Consistency. Name and required fields are in place
- 100Running it twice. Mutating operations check current state
- 100Progress reporting. Reports progress
- medium Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- low The skill ranks results itself: that belongs to the system behind the tool, not the model
- high The skill tells the model to perform an irreversible action with no human approval
Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.
Quality signals
- +5Description has no quoted example phrases that should trigger the skill
- +4Description does not say when NOT to use the skill (false activations)
- +4No input/output examples
- +1No license
- +2Single-language instructions
- +3Description length 690: enough signal without eating the budget
- +4Structure: 33 headings
- +3Step-by-step instructions: 166 items
- +3Output format is stated explicitly
- +4Reference files are cited in the instructions (2 of 2)
Quality base 70; lint remarks subtract, signals add up to 100. Result: 77.