Skillv2.0.1

ClawScan security

PharmaClaw Pharmacology Agent · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

BenignApr 2, 2026, 2:49 AM

Verdict: benign
Confidence: high
Model: gpt-5-mini
Summary: The skill's code and instructions are coherent with its stated purpose (ADME/PK profiling from SMILES); it uses RDKit locally and optionally calls a documented external ADMETlab API — no unexplained credential or install requests — but be aware that using the ADMETlab integration will send SMILES to a third‑party endpoint which may expose proprietary chemistry.
Guidance: This skill appears to do what it says: local RDKit-based ADME profiling and an optional script that posts SMILES to ADMETlab 3.0. Before installing or running: (1) Ensure RDKit and requests are available in your environment. (2) If your molecules are proprietary or confidential, avoid running admetlab3.py (it will send SMILES to a third‑party API). (3) If you want to restrict network access, run only scripts/chain_entry.py (it uses local RDKit models and does not call the remote API). (4) Review or sandbox the optional lab_hook dashboard integration if you have a local 'pharmaclaw-lab-ui' directory to avoid unintended local imports.

Review Dimensions

Purpose & Capability: noteThe name/description match the provided scripts: chain_entry.py performs RDKit-based descriptor and rule-based ADME predictions and admetlab3.py implements optional ADMETlab 3.0 API calls and a RDKit fallback. No unrelated binaries or secrets are required. Minor inconsistency: the top-level wording implies automatic combination of ADMETlab+RDKit, but chain_entry.py does not call admetlab3.py automatically — the ADMETlab integration is provided as a separate script the user can run.
Instruction Scope: noteSKILL.md instructs running the included Python scripts. chain_entry.py operates locally on a provided SMILES and does not exfiltrate other system data. admetlab3.py issues HTTP POSTs to https://admetlab3.scbdd.com/service/evaluation (documented in the code) — expected for the advertised ML integration but important to note: sending SMILES to that external service will transmit molecule structures off-host (potential IP/privacy risk). There are no instructions to read unrelated files or environment variables.
Install Mechanism: okNo install spec is provided (instruction-only install). The code relies on standard Python packages (requests) and RDKit; this is proportional to the functionality. Nothing in the skill downloads arbitrary archives or runs external installers.
Credentials: noteThe skill declares no required environment variables or credentials and the code does not attempt to read secrets. It does rely on RDKit and optional SA_Score/PAINS RDKit contribs; ensure those dependencies are available. The optional lab_hook import inserts a relative path to a neighboring 'pharmaclaw-lab-ui' directory but falls back to a no-op if not present — not a credential access but it can call a local dashboard if that package exists.
Persistence & Privilege: okalways is false and the skill does not modify other skills or system configuration. It only prints JSON output and calls an optional external API; autonomous invocation is allowed by default (normal for skills) but not combined with other high-risk requests.