Pharmacology agent for ADME/PK profiling of drug candidates from SMILES. Computes drug-likeness (Lipinski Ro5, Veber rules), QED, SA Score, ADME predictions...
Security Analysis
high confidenceThe skill's code, instructions, and requirements are coherent for an RDKit-based ADME/PK profiling agent and do not request unrelated credentials or install external software at runtime.
Name/description (ADME/PK profiling from SMILES) matches the included script and documentation. The code computes descriptors, rule-based ADME predictions, QED, SA score (if available), and PAINS via RDKit — all consistent with the stated purpose.
SKILL.md instructs running scripts/chain_entry.py with a JSON containing a SMILES string; the script consumes that input and returns a JSON report. The instructions and script do not read unrelated files, environment variables, or send data to external endpoints. The only optional behavior hinted at (resolving via PubChem) is documented as 'caller should provide SMILES' and is not implemented as a network call.
No install spec is provided (instruction-only with one included script). The script depends on RDKit and optional SA_Score/PAINS modules from RDKit's contrib area; there are no downloads from arbitrary URLs or other high-risk install steps.
The skill declares no required environment variables, credentials, or config paths. The script does append RDKit's contrib path to sys.path to try to import sascorer (local dependency) but does not access secrets or unrelated environment variables.
always is false and the skill does not modify agent/system configuration or request persistent privileges. It only runs standalone computations on supplied SMILES.
Guidance
This skill appears coherent and implements local, RDKit-based ADME profiling. Before installing or running: (1) ensure RDKit is installed from a trusted source and the runtime environment is isolated (virtualenv/container) so optional imports (SA_Score, PAINS catalog) won't unexpectedly pull or execute untrusted code; (2) verify the script's provenance (source is unknown) if you require supply-chain assurance; (3) supply SMILES locally — the skill does not perform network lookups; (4) if you need automated pipelines, confirm downstream agents (toxicology, ip-expansion) are trusted. Overall the footprint is proportional, but run untrusted code in a controlled environment.
Latest Release
v1.1.0
**Pharma Pharmacology Agent v1.1.0** - Initial production release with comprehensive ADME and drug-likeness profiling from SMILES - Implements Lipinski Rule of Five, Veber oral bioavailability rules, QED, SA Score, and PAINS filtering - Predicts BBB permeability, aqueous solubility, GI absorption, CYP3A4 inhibition risk, P-gp substrate likelihood, and plasma protein binding - Provides automated risk flagging and structured JSON output for seamless chaining - Robust error handling for invalid or missing inputs - Validated with multiple example molecules and chemistry-query integration
More by @Cheminem
Published by @Cheminem on ClawHub
