@skyeav/tablassist

OpenCode plugin for Tablassert table configuration — entity resolution, YAML validation, and Biolink documentation tools.

Installation

npm install @skyeav/tablassist

Add the plugin to your OpenCode configuration. The plugin requires the Tablassist CLI to be installed and available on PATH:

pip install tablassist

Current Tablassist CLI releases include Docling in the base install, so the plugin can call extract-text-semantic without any separate semantic extraction helper.

What It Does

This plugin gives AI agents access to the full Tablassist CLI toolset through OpenCode:

• Entity Resolution — Search and resolve biological CURIEs (genes, diseases, chemicals)
• YAML Validation — Validate Tablassert table configurations with detailed error reporting
• Config Audit — Run deep, report-first reviews of existing configs before semantic edits
• Biolink Documentation — Look up categories, predicates, and qualifiers from the Biolink model
• Data Preview — Inspect Excel and CSV files before building configurations
• Dual Text Extraction — Use fast raw Textract extraction or richer semantic Docling extraction depending on the source
• Configuration Reference — Access production examples and schema documentation

Full config validation expects template: as the top-level YAML key, with optional sections:. The standalone validate-section-str tool is only for checking an individual merged section shape.

Document extraction tools:

• extract-text for fast raw text extraction from PDFs, DOCX, and similar files
• extract-text-semantic for richer Markdown or plain-text extraction with ocr=auto by default

The semantic extractor calls the CLI's built-in Docling path directly, so the plugin exposes the same semantic extraction endpoint that the CLI ships by default.

Architecture

The plugin wraps the Tablassist CLI and exposes its commands as OpenCode tools:

plugin/
├── src/
│   ├── index.ts          # Plugin entry point
│   ├── cli.ts            # CLI runner and shell execution
│   ├── cache.ts          # Resource caching system
│   ├── hooks/            # System prompt and YAML validation hooks
│   └── tools/            # Tool definitions (API, Biolink, files, schema)
└── agents/               # Agent definitions (configurator, extractor, builder)

Three agents orchestrate the configuration workflow:

| Agent | Role | |---|---| | the-configurator | Primary orchestrator for building table configs | | the-extractor | Subagent for data extraction and preview | | the-builder | Subagent for YAML construction and validation |

Slash Commands

All commands are namespaced with the tablassist: prefix.

• /tablassist:audit <config-path> performs a deep, report-first review: validates structure, fetches the PMC publication archive when available, uses semantic extraction for structured document review, consults schema and Biolink references, and recommends improvements without applying them until you approve.
• /tablassist:validate <config-path> validates a config file and reports schema errors. If PMC-backed follow-up is needed, it should prefer download-pmc-tar, then use pmc-oa-readme as the official fallback.
• /tablassist:preview <file-path> previews rows from a CSV, TSV, or Excel file.
• /tablassist:search <term> searches for CURIE candidates matching a term.

PMC retrieval guidance for agents:

1. Prefer Tablassist-native tools and the /tablassist:* slash commands before open-web approaches.
2. For PMC content, follow the three-step retrieval chain:
   1. Try download-pmc-tar first.
   2. If that fails, call pmc-oa-readme to obtain the AWS CLI commands, then execute them via bash (e.g., aws s3 cp --no-sign-request ...).
   3. Only if both steps above fail, fall back to direct web retrieval (e.g., curl or webfetch) as a last resort.
3. Do not retry guessed PMC, S3, or publisher links with curl or similar direct-download commands after a failed PMC archive download; those links often return HTML instead of the expected archive. This prohibition applies to guessed URLs only — executing the official AWS CLI commands returned by pmc-oa-readme via bash is expected and required.

Example:

/tablassist:audit ./configs/example.yaml

Development

bun install                          # install dependencies
bun run ./src/index.ts               # run
bun x tsc --noEmit                   # type check
bun test                             # run all tests

License

Apache License 2.0