RDKit

Overview

This skill exposes lightweight local RDKit tooling through small CLI entrypoints. Every script reads a request JSON file, writes a stable result.json, and can echo the same top-level payload to stdout with --json.

The skill is fully local:

• no network access
• no ChemQA or DebateClaw runtime dependency
• structured JSON errors when RDKit is unavailable

Harbor Trial execution

harbor-agent-infra injects allowlisted skills into the disposable Harbor Trial container. OpenClaw discovers this skill at $HOME/.openclaw/skills/rdkit, and the agent working directory is /workspace. The old benchmark runner, host workspace paths, and legacy benchmark imports are not available in a Trial. Install missing provider packages inside the Trial (prefer a venv) and invoke the entrypoint directly.

python "$HOME/.openclaw/skills/rdkit/scripts/<capability>.py" \
  --request-json /workspace/requests/rdkit.json \
  --output-dir /workspace/outputs/rdkit \
  --json

--output-dir is required and is created automatically.

Capabilities

• canonicalize.py: parse SMILES or InChI, sanitize, optionally strip atom maps, and return canonicalized structure data
• descriptors.py: formula, mass, charge, donor/acceptor counts, TPSA, logP, and compact molecule summary
• functional_groups.py: curated SMARTS matches for common functional groups and reactive handles
• substructure.py: SMARTS matching against one or more molecules
• rings_aromaticity.py: ring counts, ring sizes, aromaticity, fused-ring hints, and heteroaromatic features
• stereochemistry.py: chiral centers, specified vs unspecified stereochemistry, and double-bond stereo features
• similarity.py: Morgan fingerprint similarity ranking against supplied candidates
• reaction_smarts.py: reaction SMARTS validation and product generation
• conformer_embed.py: ETKDG embedding plus an explicitly selected MMFF or UFF optimization; the request must name the force field
• conformer_mmff.py: ETKDG embedding plus MMFF optimization, with an optional MMFF94 or MMFF94s variant
• conformer_uff.py: ETKDG embedding plus UFF optimization

Execution Rules

• Canonicalize raw external structures before downstream structural analysis.
• For NMR peak-count questions, use RDKit only for structure facts such as canonicalization, formula, ring systems, and stereochemistry. Do not look for or create graph-symmetry proton-count tooling; final NMR peak counts require manual spectroscopic interpretation.
• Use conformer scripts only when 3D geometry is relevant. Select the force field from the task or scientific protocol; never infer it from parameter availability and never substitute MMFF for UFF or UFF for MMFF.
• Set num_conformers and random_seed explicitly for every conformer request. The skill does not provide defaults or preferred values for either field.
• Read routing-rules.md for script selection and references/contracts.md for request and result contracts.