Back to skill

Security audit

Lifescience Target Intelligence

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed life-sciences research workflow that gathers public and MCP-backed target intelligence, with no evidence of hidden execution, persistence, credential theft, or destructive behavior.

Before installing, confirm that external lookups to public scientific databases and web search are acceptable for your research context, because target names, gene IDs, and commercial questions may disclose areas of interest to those services. No hidden or destructive behavior was found.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (9)

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

Referencing the OpenTargets external domain directly in executable-style guidance is a true security concern because it promotes data flow to an unsanctioned external service without context-dependent validation of what is being transmitted. Even if the intended payload is scientific metadata, target names, IDs, or associated project context could still reveal confidential research interests or pipeline focus.

Content

Scanner excerpt · SKILL.md (reported line 84)May include surrounding context.

md
})

# OpenTargets — disease associations and tractability
r = requests.post("https://api.platform.opentargets.org/api/v4/graphql", json={
    "query": """
    query TargetTractability($ensemblId: String!) {
      target(ensemblId: $ensemblId) {

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

Referencing the OpenTargets external domain directly in executable-style guidance is a true security concern because it promotes data flow to an unsanctioned external service without context-dependent validation of what is being transmitted. Even if the intended payload is scientific metadata, target names, IDs, or associated project context could still reveal confidential research interests or pipeline focus.

Content

Scanner excerpt · SKILL.md (reported line 84)May include surrounding context.

md
})

# OpenTargets — disease associations and tractability
r = requests.post("https://api.platform.opentargets.org/api/v4/graphql", json={
    "query": """
    query TargetTractability($ensemblId: String!) {
      target(ensemblId: $ensemblId) {

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

Referencing the OpenTargets external domain directly in executable-style guidance is a true security concern because it promotes data flow to an unsanctioned external service without context-dependent validation of what is being transmitted. Even if the intended payload is scientific metadata, target names, IDs, or associated project context could still reveal confidential research interests or pipeline focus.

Content

Scanner excerpt · SKILL.md (reported line 84)May include surrounding context.

md
})

# OpenTargets — disease associations and tractability
r = requests.post("https://api.platform.opentargets.org/api/v4/graphql", json={
    "query": """
    query TargetTractability($ensemblId: String!) {
      target(ensemblId: $ensemblId) {

External Transmission

Medium
Category
Data Exfiltration
Confidence
88% confidence
Finding

The skill also embeds direct GET requests to RCSB PDB and PDBe, again instructing outbound calls beyond the managed tool ecosystem. While the data sought is public structural biology information, the pattern still creates a policy bypass channel and can leak sensitive internal research context through query selections and identifiers.

Content

Scanner excerpt · SKILL.md (reported line 97)May include surrounding context.

md
})

# RCSB PDB — experimental structure availability
r = requests.get("https://data.rcsb.org/rest/v1/core/entry/[pdb_id]")
# PDBe — structure quality scores
r = requests.get(f"https://www.ebi.ac.uk/pdbe/api/validation/residuewise_outlier_summary/entry/{pdb_id}")

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The markdown instructs the skill to render card grid stage labels using Chinese terms (已批准, 早期) alongside English phase labels. This imposes a specific language choice in output without offering the user a language or locale option, which matches the language-policy violation criteria.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The document requires 'Web Search validation' for pricing/commercial data even though the skill is described as an internal target-intelligence skill centered on a specific MCP server. This creates scope expansion and can cause the agent to access external sources with different trust, privacy, and policy characteristics, increasing the risk of data leakage, unvetted source use, and inconsistent outputs.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The report template mandates section labels in Chinese for multiple recommendation categories, such as '研发方向', '商务/并购', and '投资建议'. This imposes a specific language/locale in the required output structure without offering the user a choice or documenting that the skill is intentionally region-specific.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

The completion checklist requires execution of PATH 7, but no PATH 7 is defined anywhere in the reference. This kind of undefined mandatory step can drive an agent to improvise missing behavior, potentially selecting unauthorized tools or inventing a workflow to satisfy completion criteria.

Content

No source excerpt is available for this finding.