Back to skill

Security audit

neon-postgres

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marked this skill suspicious because scanners report undisclosed model-based processing and external data transfers, and some referenced content was only partially analyzed.

The verdict is suspicious, so consider the reported data transfers and scope mismatch before installing. Review whether sending database content to external cache or AI services fits your privacy requirements, and note that analysis of two referenced files was partial; the scan does not establish that those files are safe.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (16)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Reranking is related to the declared hybrid-search functionality, but this implementation crosses an explicit scope boundary: its central operation is model inference, not database querying or search configuration. The timeout, batching, validation, and sorting are supporting details and are not themselves mismatches. The mismatch is the AI invocation and associated submission of candidate content despite the description routing LLM calls elsewhere.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied chunk performs skill management and contains no database guidance or database operations. Although the description references the parent neon skill for other tasks, it does not declare setting up that skill. As supplied, this is a materially different primary behavior.

Content

No source excerpt is available for this finding.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 304)May include surrounding context.

md
- For semantic search, including generating embeddings with the Neon AI Gateway, read [Vector search](references/vector-search.md).

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 306)May include surrounding context.

md
- For combining semantic and lexical results with fusion or reranking, read [Hybrid search](references/hybrid-search.md).

Memory Manipulation

High
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.

Content

Scanner excerpt · references/branching.md (reported line 145)May include surrounding context.

neon branches reset <id|name> --parent --preserve-under-name --project-id

text

`--preserve-under-name` keeps the pre-reset state as a backup branch for rollback, but adds one extra branch to clean up later.

Optional context setup to avoid repeating `--project-id`:

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 25)May include surrounding context.

FIRST: Use the parent neon skill for a Neon overview, getting started with Neon, Neon development best practices, and more.

If the neon skill is not installed, fetch it from https://neon.com/docs/ai/skills/neon/SKILL.md or install it with:

bash
neon skills -s neon -y

External caching is recommended without warning that database contents may be sent to a third-party service.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The skill instructs the agent to cache database rows outside Postgres and explicitly suggests Upstash Redis. This can transfer application configuration, feature flags, or other database contents to another provider, but the workflow provides no privacy disclosure or guidance on selecting data appropriate for external storage.

Content

No source excerpt is available for this finding.

The caching workflow recommends a third-party Redis service without warning that database query results will be sent to and stored by that provider.

Medium
Category
Output Handling
Confidence
95% confidence
Finding

The fix instructs the agent to add caching and recommends Upstash Redis when no cache already exists; this is an implementation instruction, not merely a tool listing. Caching query results with an external provider introduces a new data destination, but the skill gives no privacy warning or instruction to check whether the data is approved for third-party storage.

Content

No source excerpt is available for this finding.

The reranking workflow sends query text and document contents to an external AI service without a privacy warning.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The supplied implementation sends the original query and candidate document titles and bodies to a hosted inference service without checking whether that processing is permitted. Its tenant filters protect retrieval authorization, but do not authorize disclosure to another processing service; sensitive documents could therefore cross a prohibited data boundary.

Content

No source excerpt is available for this finding.

The hybrid-search example performs AI model calls despite the manifest assigning LLM calls to the parent `neon` skill.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

The manifest explicitly routes LLM calls to the parent neon skill, but this example selects a Jev decision model and invokes it through the AI SDK, sending the search query and candidate titles and bodies for inference. Although reranking is relevant to hybrid search, directly implementing this paid AI Gateway operation crosses that stated boundary; the file also provides an RRF alternative that requires no additional model call.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx drizzle-kit'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx drizzle-kit'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx drizzle-kit'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx drizzle-kit'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Embedding and reranking instructions send user content to AI Gateway without a privacy warning.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The examples transmit raw document bodies and query text to AI Gateway, including bulk document processing, without a data-classification or processing-approval check. A user authorized to insert or search database records is not necessarily authorized to send their contents to an inference service, making this an unsafe default for confidential corpora.

Content

No source excerpt is available for this finding.

The embedding workflow sends document bodies and query text to Neon's AI Gateway without a privacy warning.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

Both embedding examples send original document bodies and search queries to a hosted AI endpoint, potentially processing an entire confidential corpus. The service is named openly, but the workflow lacks a prerequisite confirming that these contents may leave the application's database-processing boundary.

Content

No source excerpt is available for this finding.