Pharmaclaw Catalyst Design

Organometallic catalyst recommendation and novel ligand design for drug synthesis reactions. Recommends catalysts (Pd, Ru, Rh, Ir, Ni, Cu, Zr, Fe) for reacti...

Security Scan

The skill's code, instructions, and resource requests are internally consistent with its stated purpose (catalyst recommendation + ligand design); it runs locally, does not require credentials, and contains no evident exfiltration or unexpected network behavior — but it does require RDKit (not declared) and produces chemistry outputs that could be hazardous or have IP/legal implications.

Purpose & Capability

Name/description align with the included files: a local catalyst database, a recommender script, a chain entry point, and a ligand designer that uses RDKit. The requested functionality (recommend + design) is implemented in the shipped scripts and the database. There are no unrelated credentials, binaries, or external services required by the code.

Instruction Scope

SKILL.md and the scripts limit operations to reading the local JSON database and generating ligand variants with RDKit. The code does not call external endpoints, open sockets, or read arbitrary system files. Note: the ligand designer suggests potentially hazardous chemical substitutions (e.g., As/arsine replacements) and reaction conditions (pressures, reactive metals) — these are domain-risk issues (lab safety / dual-use) rather than software exfiltration, but users should treat chemical recommendations carefully and validate before any lab work.

Install Mechanism

There is no install spec (instruction-only + bundled scripts). That is low risk from a code-fetching perspective. However, the ligand_designer.py has an unconditional RDKit import and exits with an error if RDKit is not installed; RDKit is therefore an undeclared runtime dependency. The skill does not auto-install RDKit or any other package, and contains no download/extract behavior.

Credentials

The skill requests no environment variables, credentials, or config paths. All data access is local (the included catalyst_database.json). No broad or unrelated secrets are requested, so environment access is proportional to the stated purpose.

Persistence & Privilege

Skill flags (always, disable-model-invocation) are default/benign; always:false. The skill does not modify other skills or system config. It runs only when invoked and does not request permanent presence or elevated agent-wide privileges.

Assessment

What to consider before installing: - The skill runs locally and uses only the bundled JSON DB and the three scripts; there are no network calls or credential requests in the code as provided. - You must install RDKit yourself for the ligand designer to work (the package is not declared in metadata nor is there an install step). If you install RDKit, do so from a trusted source (conda or the upstream project) because RDKit wheels/conda packages are sizable and complex. - Review the code and outputs before using any recommendations in the lab. The designer suggests chemical substitutions (including arsenic-containing ligands) and reaction conditions that can be hazardous — these are domain-safety concerns, not software security issues. - The skill references "IP Expansion" and patentability in SKILL.md but does not implement patent-checking; if you plan to use this for inventions or filings, validate any legal/IP claims with counsel. - Source provenance is unknown (no homepage). If provenance or license matters to you, ask the publisher for origin and licensing; otherwise treat the code as untrusted third-party code and audit accordingly.

Catalyst Design Agent v1.0.0

Overview

Recommends organometallic catalysts for drug synthesis steps and designs novel ligand modifications. Two core workflows: recommend (find the right catalyst) and design (create novel ligand variants).

Quick Start

# Recommend catalysts for a Suzuki coupling
python scripts/catalyst_recommend.py --reaction suzuki

# Recommend with constraints (prefer cheap, earth-abundant)
python scripts/catalyst_recommend.py --reaction "C-N coupling" --constraints '{"prefer_earth_abundant": true, "max_cost": "medium"}'

# Design novel ligand variants from PPh3
python scripts/ligand_designer.py --scaffold PPh3 --strategy all --draw

# Full chain: reaction → catalyst → ligand optimization
python scripts/chain_entry.py --input-json '{"reaction": "suzuki", "context": "retrosynthesis"}'

Scripts

scripts/catalyst_recommend.py

Scores and ranks catalysts from curated database (12 catalysts, 28 reaction types).

--reaction <type>              Required. e.g., suzuki, metathesis, C-N coupling, hydrogenation
--substrate <SMILES>           Optional. Substrate context
--constraints <JSON>           Optional. {prefer_metal, max_cost, prefer_earth_abundant}
--enantioselective             Flag. Prioritize chiral catalysts

Scoring (0-100): reaction match (50), cost (15), metal preference (10), enantioselectivity (10), loading efficiency (5), advantages (5), earth-abundance (5).

scripts/ligand_designer.py

Generates novel ligand variants via three strategies:

Strategy	Method	Output
steric	Add methyl/iPr/tBu to aromatic rings	Modified SMILES + properties
electronic	Add OMe/F/CF3 substituents	Modified SMILES + properties
bioisosteric	P→NHC, phenyl→pyridyl, phosphine→phosphite	Conceptual suggestions + rationale

--scaffold <SMILES|name>       Required. PPh3, NHC_IMes, NHC_IPr, PCy3, dppe, dppp, or raw SMILES
--strategy <type>              steric|electronic|bioisosteric|all (default: all)
--draw                         Generate 2D grid PNG of variants
--output <path>                Save JSON results to file

scripts/chain_entry.py

Standard PharmaClaw chain interface. Accepts JSON, routes to recommend/design/both.

Input keys: reaction, scaffold/ligand, substrate/smiles, constraints, enantioselective, strategy, draw, context

If only reaction is given, also auto-runs ligand optimization on the top recommended catalyst's ligand.

Chaining

From	Input	To	Output
Chemistry Query / Retrosynthesis	Reaction type needed for a synthesis step	Catalyst Design	Ranked catalysts + conditions
Catalyst Design	Top catalyst ligand SMILES	Ligand Designer	Novel ligand variants
Catalyst Design	Novel ligand SMILES	IP Expansion	Patent landscape check
Catalyst Design	Recommended conditions	Chemistry Query	Forward reaction simulation

Database

references/catalyst_database.json — 12 catalysts, 8 metals, 28 reaction types. Includes SMILES, conditions, loading ranges, cost ratings, advantages/limitations, and literature DOIs.

Expandable: add entries following the existing schema.

Ligand Aliases

PPh3, PCy3, dppe, dppp, NHC_IMes, NHC_IPr — resolved automatically to SMILES.