Name: chemical_index
Author: zgraper

Imported from zgraper/chemical_index (AGENTS.md). Install upstream with npx skills add zgraper/chemical_index. Copyright stays with the author.

The concept is that Cornbelt AI will be able to pull information from labels and give it to farmers. We figure the best way to do this is create an index, then pull the label if there is a query regarding it.

Project Architecture

This is a single Python package (chemical_index) installed via pip install -e .. All commands are exposed through the chemical-index CLI entry point.

Module Map

Module	Purpose
`schema.py`	SQLite schema creation (`product_versions`, `index_runs` tables) and migrations
`normalize.py`	Canonicalises raw source dicts into structured product records
`hashing.py`	SHA-256 content hashing for change detection in sync runs
`build_index.py`	Non-destructive full rebuild of the index from a JSON source
`sync_index.py`	Incremental sync – only inserts new version rows when data has changed
`search.py`	Five search modes: `epa_reg_no`, `product_name`, `fuzzy`, `active_ingredient`, `registrant`
`retrieval.py`	Evaluation harness – runs test cases and reports top-1/top-3 accuracy and MRR
`label_retrieval.py`	Fetches PDF labels (with local caching) and extracts structured sections
`pdf_parser.py`	`pypdf`-based text extraction and whitespace normalisation
`section_extractor.py`	Rule-based extraction of labelled sections (directions for use, PPE, REI, etc.)
`validate.py`	Database integrity checks (duplicate `is_latest` rows, orphan run IDs, duplicate hashes)
`safety.py`	Strips recommendation-style language and appends a regulatory disclaimer
`api.py`	FastAPI app exposing `/search`, `/product/{epa_reg_no}`, `/label/{epa_reg_no}`, `/label/{epa_reg_no}/sections`
`cli.py`	Click CLI: `build-index`, `sync-index`, `search`, `evaluate`, `validate`, `serve`, `extract-label`, `demo`

Key Data Flows

Ingestion: build_index / sync_index → normalize → hash → SQLite product_versions
Search: CLI/API → search.py → SQLite query → ranked result list with explain field
Label retrieval: label_retrieval.py → database lookup → PDF download (cached) → pdf_parser → section_extractor → structured sections dict
Evaluation: retrieval.py reads JSON test cases → runs search → computes MRR/accuracy metrics → exports JSON + CSV reports

Test Infrastructure

All tests are under tests/ and use pytest with tmp_path fixtures (no external services needed)
tests/data/sample_products.json – five sample EPA-registered products (also copied to data/products.json for CLI quickstart)
tests/data/test_cases.json – retrieval golden test cases for those five products
FastAPI tests use httpx via TestClient with a monkeypatched CHEMICAL_INDEX_DB env var
PDF-dependent tests are skipped or mocked so the suite runs without real PDF files

Adding New Source Data

Replace or extend data/products.json with real EPA product records, then run:

chemical-index build-index --source data/products.json --db index.sqlite

For ongoing updates use sync-index so previous version history is preserved.

Install chemical_index and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

{ "ocm": "1", "id": "zgraper-chemical-index-chemical-index", "kind": "skill", "name": "chemical_index", "description": "The concept is that Cornbelt AI will be able to pull information from labels and give it to farmers. We figure the best way to do this is create an index, then pull the label if there is a query regarding it.", "publisher": "zgraper", "version": "1.0.0", "capabilities": { "domains": [ "data_analysis" ], "tags": [ "skill-md", "github-agents" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "The concept is that Cornbelt AI will be able to pull information from labels and give it to farmers. We figure the best way to do this is create an index, then pull the label if there is a query regarding it." ], "primary": false, "metadata": { "source": { "provider": "github-agents", "repository": "https://github.com/zgraper/chemical_index", "path": "AGENTS.md", "ref": "7672f9adf373346d1d83937df495027b354b0161", "url": "https://github.com/zgraper/chemical_index/blob/7672f9adf373346d1d83937df495027b354b0161/AGENTS.md", "key": "zgraper/chemical_index/AGENTS.md" } }, "instructions": "The concept is that Cornbelt AI will be able to pull information from labels and give it to farmers. We figure the best way to do this is create an index, then pull the label if there is a query regarding it.\n\n## Project Architecture\n\nThis is a single Python package (`chemical_index`) installed via `pip install -e .`. All commands are exposed through the `chemical-index` CLI entry point.\n\n### Module Map\n\n| Module | Purpose |\n|--------|---------|\n| `schema.py` | SQLite schema creation (`product_versions`, `index_runs` tables) and migrations |\n| `normalize.py` | Canonicalises raw source dicts i", "cost": { "context_tokens": 781 } }

chemical_index

About

Project Architecture

Module Map

Key Data Flows

Test Infrastructure

Adding New Source Data

Install chemical_index and it becomes one more candidate the router can pick - when it fits.

Use it

Manifest

Reviews