name	biocheminsight-text-models
description	Use when a BioChemInsight language model needs to normalize assay markdown into JSON or extract a final compound ID from free-form textual reasoning or OCR output.
model	inherit
effort	high
context	auto

BioChemInsight Text Models

Use this skill for runtime text-model tasks.

This skill inherits shared runtime conventions from biocheminsight-model-common and only overrides text-task-specific behavior.

Scope

Assay markdown/table extraction into strict JSON dictionaries
Compound-ID extraction from free-form text or model reasoning
Stable parsing rules for downstream machine-readable outputs

Runtime rules

Output machine-readable JSON only when the caller asks for JSON.
Never hallucinate compound IDs outside the provided allowlist.
Represent assay cells as rich objects: value, unit, method, description, confidence, reason.
Keep prompt contracts short; use examples and schemas for detailed shape.
Prefer returning "None" over speculative IDs.
For page-detection tasks, emit exactly one decisions item per provided page.
For assay extraction, preserve method/unit/symbol descriptions in the rich assay object so exports do not lose scientific context.
Treat continuation context as read-only table metadata. It may explain headers, units, methods, and assay ownership for a current page/table chunk, but output records must come only from the current chunk.
Use a planning step for multi-page assay extraction when page/table continuation is ambiguous. The planner decides context inheritance; extractor prompts execute that plan without inventing records.
If a bioactivity page overlaps a structure page, allow the planner to choose a visual-structure anchor strategy. Text extraction must not invent compound IDs; visual anchoring should be handled by a bounded harness step with source page and structure-record provenance.

Runtime references

references/content_to_dict_prompt.md
references/content_to_multi_assay_dict_prompt.md
references/detect_assay_pages_prompt.md
references/plan_assay_extraction_context_prompt.md
references/get_compound_id_from_description_prompt.md
references/resolve_compound_id_alias_prompt.md
references/runtime.json
references/output_schemas.json

BioChemInsight Text Models

Use this skill for runtime text-model tasks.

This skill inherits shared runtime conventions from biocheminsight-model-common and only overrides text-task-specific behavior.

Scope

Assay markdown/table extraction into strict JSON dictionaries

Compound-ID extraction from free-form text or model reasoning

Stable parsing rules for downstream machine-readable outputs

Runtime rules

Output machine-readable JSON only when the caller asks for JSON.

Never hallucinate compound IDs outside the provided allowlist.

Represent assay cells as rich objects: value, unit, method, description, confidence, reason.

Keep prompt contracts short; use examples and schemas for detailed shape.

Prefer returning "None" over speculative IDs.

For page-detection tasks, emit exactly one decisions item per provided page.

For assay extraction, preserve method/unit/symbol descriptions in the rich assay object so exports do not lose scientific context.

Treat continuation context as read-only table metadata. It may explain headers, units, methods, and assay ownership for a current page/table chunk, but output records must come only from the current chunk.

Use a planning step for multi-page assay extraction when page/table continuation is ambiguous. The planner decides context inheritance; extractor prompts execute that plan without inventing records.

If a bioactivity page overlaps a structure page, allow the planner to choose a visual-structure anchor strategy. Text extraction must not invent compound IDs; visual anchoring should be handled by a bounded harness step with source page and structure-record provenance.

Runtime references

references/content_to_dict_prompt.md

references/content_to_multi_assay_dict_prompt.md

references/detect_assay_pages_prompt.md

references/plan_assay_extraction_context_prompt.md

references/get_compound_id_from_description_prompt.md

references/resolve_compound_id_alias_prompt.md

references/runtime.json

references/output_schemas.json

biocheminsight-text-models