Back to skill

Security audit

Daeva

Security checks for vulnerabilities and agentic risk

Overview

The skill’s Daeva purpose is coherent, but it needs review because it can route user files to local or remote services and perform shared pod/package management without enough confirmation safeguards.

Review this skill before installing. Use it only with a Daeva instance you trust, confirm whether DAEVA_URL points to a remote host before sending files, avoid installing pod packages from untrusted sources, and be careful with start/stop/swap actions on shared GPU infrastructure.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T08 · Insecure Dependencies

Error
Location
SKILL.md:143
Finding

Unverified Pod Packages Can Execute Installation Hooks

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:82
Finding

Unquoted Environment-Controlled Base URL in Curl Commands

Content
View full analysis
# Job state curl -s $DAEVA_BASE/jobs//result # Job result when complete curl -s $DAEVA_BASE/jobs # List all jobs ``` ```bash # List all registered pods and their runtime state curl -s $DAEVA_BASE/pods # Register a new pod from a manifest curl -s -X POST $DAEVA_BASE/pods/register \ -H 'Content-Type: application/json' \ -d '{ ... pod manifest JSON ... }' # Install a pod package by alias (e.g. "whisper") curl -s -X POST $DAEVA_BASE/pods/create \ -H 'Content-Type: application/json' \ -d '{"alias":"whisper"}' # List available aliases from the registry curl -s $DAEVA_BASE/pods/aliases # List already-installed packages curl -s $DAEVA_BASE/pods/installed ``` The base URL is obtained from environment-controlled values: ```bash DAEVA_BASE="${DAEVA_URL:-http://127.0.0.1:${DAEVA_PORT:-8787}}" ``` ### Technical Analysis Although the initial health-check example quotes the base URL, later curl examples expand `$DAEVA_BASE` without double quotes. In a POSIX-style shell, an unquoted variable is subject to field splitting and pathname expansion. If its value contains whitespace, one logi ...[truncated 2666 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger conditions are overly broad and instruct the agent to invoke this skill for generic requests like transcription, OCR, image generation, and vision, even when the user does not mention Daeva. That creates a routing vulnerability where ordinary requests may be redirected to a local or remote shared service, causing unintended external processing, privacy exposure, or infrastructure-side effects without clear user intent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill exposes commands to register, install, activate, stop, and swap pods on a shared service, but does not require confirmation or warn that these actions may disrupt other users. In a shared infrastructure context, allowing routine execution of lifecycle actions can interrupt running workloads, alter system state, or install untrusted packages with operational and security consequences.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

The skill instructs sending local files and prompts to a resolved Daeva base URL that may point to a remote host via DAEVA_URL, which creates a real data exfiltration path. Because the skill handles sensitive inputs like audio, images, OCR documents, and vision requests, broad routing plus remote endpoint support increases the risk of transmitting user data off-host or to an unintended shared service.

Content

Scanner excerpt · SKILL.md (reported line 78)May include surrounding context.

bash
# Transcribe audio
curl -s -X POST $DAEVA_BASE/jobs \
  -H 'Content-Type: application/json' \
  -d '{"type":"transcribe-audio","files":[{"source":"path","path":"/tmp/audio.wav"}]}'