Back to skill

Security audit

Gemini Notebook / NotebookLM: podcast, Video Overview, slides (useapi.net)

Security checks for vulnerabilities and agentic risk

Overview

The scanners describe a skill for using the useapi.net Gemini Notebook API to create content; the verdict is benign, with notes for additional output types beyond those described and incomplete scan coverage.

The verdict is benign, not a guarantee of zero risk. Before installing, note that the skill can generate more types of content than its description lists, and check that you are comfortable sending any selected files to the useapi.net service. The scan was partial: one referenced helper was unavailable, and tool-misuse analysis hit a parser limit.

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 Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The primary behavior matches the declared purpose, but the script also exposes additional artifact-generation capabilities. TYPE is passed directly into the API request without restriction, and the comments explicitly identify supported output types outside the description. The asynchronous submission and job-ID logging are supporting implementation details, not mismatches. The sourced helper is not supplied, so its behavior cannot be assessed.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · scripts/_common.sh (reported line 6)May include surrounding context.

sh
# Written for bash 3.2 and later (the macOS default), so no bash 4 features.
set -euo pipefail

API="https://api.useapi.net/v1/gemini-notebook"
: "${USEAPI_TOKEN:?Set USEAPI_TOKEN to your useapi.net API token (https://useapi.net/docs/start-here/setup-useapi)}"
command -v jq > /dev/null || { echo "jq is required (https://jqlang.org/download/)" >&2; exit 1; }

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% 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 · scripts/_common.sh (reported line 13)May include surrounding context.

sh
enc() { jq -rn --arg v "$1" '$v|@uri'; }
log() { echo "$(date +%H:%M:%S) $*" >&2; }

# acurl [curl args ...] -> curl with the Authorization header read from stdin (-H @-),
# so the token does not show up in the process list
acurl() { curl "$@" -H @- <<< "Authorization: Bearer $USEAPI_TOKEN"; }

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% 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 · scripts/upload-file.sh (reported line 29)May include surrounding context.

sh
OUT=$(mktemp)
trap 'rm -f "$OUT"' EXIT
CODE=$(acurl -sS --max-time 600 -o "$OUT" -w '%{http_code}' -H "Content-Type: $TYPE" --data-binary "@$FILE" \
  "$API/sources/upload?notebook=$(enc "$NOTEBOOK")&name=$(enc "$(basename "$FILE")")") || true
if [ "${CODE:0:1}" != "2" ]; then
  echo "POST /sources/upload -> HTTP ${CODE:-000}: $(cat "$OUT")" >&2

The generator supports additional artifact types beyond the audio, video, and slide outputs declared in the manifest.

Low
Category
Excessive Agency
Confidence
96% confidence
Finding

The script accepts an unrestricted artifact type and submits it to the artifacts API without checking that it is audio, video, or slides. Lines L08–L09 explicitly advertise reports, quizzes, flashcards, infographics, and mind maps, confirming an intentional expansion beyond the manifest's stated output scope.

Content

No source excerpt is available for this finding.