rdkit-agent Skill (scottmreed)

Overview

The rdkit-agent Skill is a Markdown instruction resource located at skills/rdkit-agent/SKILL.md in the scottmreed/rdkit-agent repository. It guides an assistant through using the separate rdkit-agent CLI, which the source describes as powered by RDKit WASM. The Skill is not itself a chemistry engine or an independently running agent; its role is to establish tool-use conventions for validation, conversion and molecular analysis.

Its central workflow is to check chemical notation before passing it to downstream operations. Inputs include SMILES, SMIRKS, conversion requests and molecule collections, with JSON payloads recommended for programmatic exchange. The instructions describe checking overall_pass, consulting corrected_values or fix_suggestions, and limiting returned fields or search results. Outputs from the underlying CLI include structured validation results, molecular descriptors, search and similarity results, converted notation and molecular drawings. The document also explains automatic JSON output outside a terminal and exit codes that distinguish validation, usage and RDKit errors.

The command reference spans fingerprints, scaffolds, functional groups, filtering, molecular editing, reaction application and other operations. Additional sections describe the CLI's MCP server mode and Node.js API; these are interfaces of the underlying package, not separate capabilities executed by the Skill. A documented boundary is that stereoisomer enumeration is unavailable in the standard WASM build and returns a structured unsupported-feature error. The instructions also highlight malformed notation and language-model formatting artifacts. Their examples provide workflow guidance, not evidence of independently verified chemical results or compatibility.

Key Features

• Validation-first instructions for SMILES and SMIRKS, including inspection of `overall_pass`, `corrected_values` and `fix_suggestions` before downstream use.
• Guidance on JSON payloads, automatic JSON output in non-terminal contexts, explicit output formatting and machine-readable exit codes.
• A command reference covering notation conversion, descriptors, functional groups, fingerprints, similarity, substructure search, scaffolds and molecular filtering.
• Output-size controls through selected descriptor fields, similarity result counts and substructure match limits.
• Examples of common chemical-string mistakes, including names or formulas substituted for SMILES, formatting artifacts and incomplete structural notation, with an automatic repair option.
• Instructions for the underlying CLI's MCP and Node.js interfaces, alongside an explicit WASM limitation for stereoisomer enumeration.

Use Cases

• Suggested evaluation: use the Skill to guide an assistant through validating generated SMILES, inspecting suggested corrections and withholding invalid strings from later analysis.
• Suggested evaluation: build a structured molecule-triage workflow that requests selected descriptors, applies filters and returns bounded similarity or substructure results.
• Suggested evaluation: compare CLI and MCP-based interaction paths for a chemistry assistant while checking error handling and unsupported-feature behavior.

How to Use

1. Read the named Skill file and distinguish its instructions from the separate rdkit-agent package. Confirm that the host assistant supports the described tool-access workflow.
2. Follow the file's availability check and workspace-local installation guidance if the CLI is absent. Treat its sandbox and network assumptions as conditions to verify in your environment, rather than guaranteed access.
3. Begin with a small, known chemical-string example. Follow the validation-first section and inspect overall_pass and any correction guidance before requesting further operations.
4. Choose a documented analysis or conversion operation. Use the JSON-input guidance and restrict descriptor fields or result counts where appropriate; inspect both the structured output and exit status.
5. Consult the WASM limitations and interface sections before evaluating stereochemistry, MCP integration or the Node.js API. Record observed behavior separately from the source's claims; the Skill itself does not establish scientific accuracy.