Back to skill

Security audit

deja-search

Security checks for vulnerabilities and agentic risk

Overview

The scanners rate this skill suspicious because it searches broadly across coding histories and its fallback can download and run an unpinned package without separate approval.

Review the skill's broad history-search scope and its package-install fallback before installing. Consider whether you are comfortable with automatic package execution and cross-agent persistent rule changes; the scan is partial because two referenced local paths were missing, so evidence is incomplete.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (4)

Activation conditions extend to nearly every coding task and concrete name, rather than requests that need historical recall.

Medium
Category
Trigger Abuse
Confidence
97% confidence
Finding

The skill mandates history searches for nearly every task and newly encountered concrete name, while searching across local coding histories with the current project merely prioritized rather than exclusively selected. This can unnecessarily introduce unrelated or sensitive cross-project material into the active agent context and increase exposure to transcript-based prompt injection, although the explicit reference-data and confirmation safeguards reduce that risk.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @vshulcz/deja-vu'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The fallback runs an npm package without an exact version, allowing the code executed by the agent to change independently of the reviewed skill. A compromised publisher account or malicious release could execute code with the agent's local permissions, including access to source code and session histories; this is a CLI supply-chain issue, not specifically an MCP-server issue.

Content

No source excerpt is available for this finding.

The fallback silently authorizes fetching and executing an npm package.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

When the binary is unavailable, the skill instructs the agent to use npx with automatic confirmation, which can download and execute package code without a separate installation approval. This crosses from searching existing local history into acquiring executable software and overlaps with RP1, but the missing authorization boundary remains a problem even if the version is pinned.

Content

No source excerpt is available for this finding.

The search skill also directs modification and propagation of global agent rules.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The manifest describes searching past coding sessions to recover prior decisions and preferences. L79 goes beyond retrieval by directing the agent to append persistent rules and optionally run deja rules sync, which copies them into every installed agent's global rules file. Requiring explicit user confirmation reduces the risk of unintended changes, but this cross-agent configuration capability is still outside the manifest's stated search purpose.

Content

No source excerpt is available for this finding.