| name | chemistry-research |
| description | Use when the work involves molecular structures, reactions, spectra, computational chemistry, crystallography or chemical databases. Covers SMILES/InChI round-trips, RDKit standardisation and fingerprints, NMR/IR/MS interpretation workflows, DFT method and basis-set choice, CIF and R-factors, PubChem/ChEMBL/CSD, and the characterisation standard a journal expects. Hazard, waste and institutional review are part of the procedure.
|
| version | 1.0.0 |
| author | Remedy |
| tags | ["research","chemistry","cheminformatics","spectroscopy","dft","safety"] |
| requires | [] |
| tools | ["lit_search","lit_fetch","cite_add","cite_check","analysis_env","analysis_run","analysis_ledger","data_profile","stats_effect_size","manuscript_build","skill_activate","file_read","file_write"] |
| triggers | ["\\b(1H NMR|13C NMR|\\bNMR\\b spectr\\w+|IR spectr\\w+|mass spectrometry|\\bHPLC\\b|GC-?MS|LC-?MS|\\bXRD\\b)\\b","\\b(reaction (?:yield|scheme|conditions)|stoichiometr\\w+|molar (?:ratio|mass)|equivalents of|reflux|recrystalli[sz]\\w+|column chromatograph\\w+)\\b","\\b(SMILES string|InChI|PubChem|CAS (?:number|registry)|RDKit|DFT calculation|B3LYP|basis sets?)\\b","\\b(safety data sheet|\\bSDS\\b sheet|COSHH|fume hood|chemical hygiene plan|waste (?:stream|disposal))\\b"] |
Chemistry research
Activate research-method first: skill_activate(skill="research-method").
It owns question framing, evidence standards, citation honesty and how to say
"we do not know". Do not restate it. This pack is what chemistry does
differently.
Scope and safety, stated once
You do the ordinary published chemistry: literature search, structure and
reaction handling, spectral interpretation, computational chemistry,
crystallography, purification and characterisation, statistics on assay data,
and the hazard assessment that goes with all of it.
You decline synthesis routes, scale-up detail, and procurement or
precursor-sourcing guidance for explosives, chemical weapons, drugs of abuse
and scheduled or watched precursors — including "hypothetical" and
"for a novel" framings, and including the last missing step of an otherwise
public route. When a request lands there, say so plainly, point at the
institutional route (EHS office, chemical hygiene officer, controlled
substances licensing, the relevant national scheduling authority), and
continue with the legitimate part of the work.
Routine hazard work stays in scope and belongs in the procedure, not a
footer: pull the SDS for every reagent before writing a step, record
GHS/hazard statements, name incompatibilities (oxidiser/fuel, acid/cyanide,
water-reactive, peroxide-formers with an expiry date), state containment
(fume hood, glovebox, blast shield), quench and waste stream per reagent, and
the PPE. See references/lab-safety-and-scope.md.
Structures before anything else
A structure that does not round-trip is a bug you will not notice later.
Canonicalise once, keep the canonical form, and record what you did:
- SMILES is not unique unless canonicalised, and canonical form is
toolkit-specific — say which toolkit and version produced it.
- InChI / InChIKey is the identity key for exact-match lookup. The
standard InChI deliberately loses information (tautomers are normalised,
some stereo layers can be absent); an InChIKey collision is a match on the
skeleton layer, not proof of identity.
- Strip salts and solvates to a parent for comparison, but keep the original
— the salt form changes the assay result.
- Check stereocentres and double-bond geometry survived; unspecified stereo
is different from racemic, and both differ from a single enantiomer.
- Charge, protonation state and tautomer are pH-dependent choices, not
facts. State the pH you assumed.
Details and the RDKit standardisation pipeline:
references/structure-representations.md, references/cheminformatics-rdkit.md.
Running chemistry code
RDKit, Open Babel, psi4, ASE and pymatgen are not importable inside Remedy —
the sidecar excludes the numeric stack. to see
what the project has, then run every calculation through in
the project's own interpreter so the script, its inputs and its outputs are
hashed into the ledger. A descriptor table with no run behind it cannot be
checked later.
