Back to skill

Security audit

SAR Analysis

Security checks for vulnerabilities and agentic risk

Overview

The skill has a coherent SAR analysis purpose, but its bundled RCSB helper is broader than disclosed because it can make arbitrary HTTP requests and optionally write raw responses to caller-chosen paths.

Review this skill before installing. It is designed for legitimate SAR analysis and discloses its SciMiner and RCSB network use, but only run it in an environment where arbitrary outbound requests and local file writes are controlled or sandboxed. Do not use it with sensitive datasets unless the REST helper is constrained to approved SciMiner/RCSB endpoints and a safe output directory.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (9)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill text claims a bounded SAR workflow, but the detected behavior indicates generic REST access, custom headers/payloads, arbitrary base URLs/paths, generic parsing, and optional raw-response file writes. That mismatch is dangerous because it can be used as a disguised general-purpose network proxy and file writer, enabling secret-bearing requests, retrieval of untrusted content, and persistence of sensitive responses outside the expected scientific workflow.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The URL construction permits absolute http/https paths to override the provided base URL, enabling arbitrary outbound requests to any reachable host. In an agent environment, this can be abused for SSRF-like access, unintended data transfer, or using the skill as a generic network client unrelated to its declared SAR-analysis role.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This code implements a fully generic HTTP client that can issue GET/POST requests to arbitrary URLs, with attacker-controlled headers, params, and bodies, then returns the response content. That capability materially exceeds a SAR-analysis skill’s stated purpose and creates a reusable exfiltration/proxy primitive if an upstream agent or prompt can influence the payload.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill requires network access and explicitly instructs reading remote Markdown docs plus saving template PDB files, but it declares no permission or allowed-tools scope. That creates an overbroad execution surface where the agent may use unintended networking or file-writing capabilities without policy-level restriction, increasing the risk of data exfiltration, SSRF-like behavior, or unsafe file operations if the skill or upstream inputs are manipulated.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · rcsb-pdb-skill/SKILL.md (reported line 8)May include surrounding context.

md
## Operating rules
- Use `scripts/rest_request.py` for all RCSB PDB and Search API calls.
- Use `base_url=https://data.rcsb.org/rest/v1` for core metadata, `https://search.rcsb.org/rcsbsearch/v2` for Search API, and `https://www.rcsb.org` for FASTA downloads.
- Core entry or assembly lookups usually do not need `max_items`; Search API results are better with query pager rows around `10` and `max_items=10`.
- Re-run requests in long conversations instead of relying on older tool output.
- Treat displayed `...` in tool previews as UI truncation, not literal request content.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · rcsb-pdb-skill/SKILL.md (reported line 23)May include surrounding context.

md
## Operating rules
- Use `scripts/rest_request.py` for all RCSB PDB and Search API calls.
- Use `base_url=https://data.rcsb.org/rest/v1` for core metadata, `https://search.rcsb.org/rcsbsearch/v2` for Search API, and `https://www.rcsb.org` for FASTA downloads.
- Core entry or assembly lookups usually do not need `max_items`; Search API results are better with query pager rows around `10` and `max_items=10`.
- Re-run requests in long conversations instead of relying on older tool output.
- Treat displayed `...` in tool previews as UI truncation, not literal request content.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · rcsb-pdb-skill/SKILL.md (reported line 34)May include surrounding context.

md
## Operating rules
- Use `scripts/rest_request.py` for all RCSB PDB and Search API calls.
- Use `base_url=https://data.rcsb.org/rest/v1` for core metadata, `https://search.rcsb.org/rcsbsearch/v2` for Search API, and `https://www.rcsb.org` for FASTA downloads.
- Core entry or assembly lookups usually do not need `max_items`; Search API results are better with query pager rows around `10` and `max_items=10`.
- Re-run requests in long conversations instead of relying on older tool output.
- Treat displayed `...` in tool previews as UI truncation, not literal request content.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The raw output save helper writes arbitrary HTTP response data to a caller-controlled filesystem path, creating an unnecessary file-write primitive outside the skill’s stated scope. Even without code execution, this can expose sensitive data on disk, overwrite application files writable by the process, or create persistence/artifact leakage in shared environments.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Saving full HTTP responses to disk without prominent disclosure increases the risk of silent retention of sensitive or proprietary remote data. In a scientific-analysis skill context, responses may include research inputs, credentials in error payloads, or other confidential records that users would not expect to be persisted locally.

Content

No source excerpt is available for this finding.