Back to skill

Security audit

Robotaxi Tracker

Security checks for vulnerabilities and agentic risk

Overview

The skill is a narrowly scoped Robotaxi Tracker helper that makes disclosed live web requests, with a minor local temporary-file safety issue to be aware of.

Install only if you are comfortable with the agent making live requests to Robotaxi Tracker and public Convex backend hosts. Prefer updating the workflow to use mktemp-created private directories before running it on shared machines.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:46
Finding

Predictable Temporary Paths Allow Local File Overwrite

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 46–72
Vulnerability Type: Unsafe temporary-file handling
Risk Level: Medium

Vulnerable Code

bash
curl -sS https://robotaxitracker.com/ -o /tmp/robotaxi_home.html
bash
grep -o '/_next/static/chunks/[^"?]*\.js' /tmp/robotaxi_home.html | sort -u > /tmp/robotaxi_chunks.txt
cat /tmp/robotaxi_chunks.txt
bash
mkdir -p /tmp/robotaxi_js
while read -r u; do
  [ -n "$u" ] || continue
  f=$(basename "$u")
  curl -sS "https://robotaxitracker.com$u" -o "/tmp/robotaxi_js/$f"
done < /tmp/robotaxi_chunks.txt
bash
rg -o '[A-Za-z0-9-]+\.convex\.cloud' -n /tmp/robotaxi_js/*.js | sort -u > /tmp/robotaxi_hosts.txt
cat /tmp/robotaxi_hosts.txt

Technical Analysis

The documented workflow repeatedly writes to fixed, predictable paths in the shared /tmp namespace. It neither creates a private temporary directory securely nor validates the ownership and file type of existing paths.

Shell output redirection and curl -o normally follow symbolic links. On a multi-user system, a local attacker may pre-create /tmp/robotaxi_home.html, /tmp/robotaxi_chunks.txt, /tmp/robotaxi_hosts.txt, or files under /tmp/robotaxi_js as symbolic links to other files writable by the account running the Agent. The subsequent workflow can then truncate or replace those link targets.

Using mkdir -p /tmp/robotaxi_js does not mitigate this issue because an attacker may create that path before the workflow begins or influence entries beneath it. The downloaded bundle filename is reduced with basename, which limits direct path traversal from the URL, but does not prevent attacks involving pre-created symbolic links at the resulting predictable destination.

Attack Path

  1. A local attacker identifies that an Agent will invoke this Skill on a shared host.
  2. Before invocation, the attacker creates a symbolic link at a predictabl ...[truncated 1251 chars]
Remediation
View remediation

Remediation Suggestions

Create a unique private temporary directory for every invocation, apply restrictive permissions, verify command success, and remove the directory when processing completes:

bash
tmpdir=$(mktemp -d) || {
  echo "Failed to create a temporary directory" >&2
  exit 1
}
chmod 700 "$tmpdir"
trap 'rm -rf -- "$tmpdir"' EXIT HUP INT TERM

home_file="$tmpdir/robotaxi_home.html"
chunks_file="$tmpdir/robotaxi_chunks.txt"
hosts_file="$tmpdir/robotaxi_hosts.txt"
js_dir="$tmpdir/js"

mkdir -m 700 -- "$js_dir" || exit 1

curl --fail --silent --show-error \
  https://robotaxitracker.com/ \
  -o "$home_file" || exit 1

Replace every fixed /tmp/robotaxi_* reference with a path beneath $tmpdir. Additional hardening should include:

  • Abort if secure temporary-directory creation fails.
  • Quote all path variables and use -- where supported.
  • Check the exit status of curl, parsing commands, and file operations.
  • Avoid reusing temporary data across invocations.
  • Do not operate with elevated OS privileges unless strictly necessary.
  • Where supported, prevent following symbolic links or create output files atomically with exclusive-create semantics.
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 (5)

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 · SKILL.md (reported line 111)May include surrounding context.

while read -r host; do [ -n "$host" ] || continue echo "=== $host ===" curl -sS "https://$host/api/query" -H 'Content-Type: application/json' --data '{"path":"queries/serviceAreas:list","args":[{"provider":"waymo"}]}' echo done < /tmp/robotaxi_hosts.txt

text

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 · SKILL.md (reported line 151)May include surrounding context.

Homepage aggregate:

bash
curl -sS https://<convex-host>/api/query \
  -H 'Content-Type: application/json' \
  --data '{"path":"queries/fleet:getHomepageData","args":[{"provider":"waymo"}]}'

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The file includes Chinese trigger examples and later defines a default output table with Chinese headers and totals, but it does not say this is optional or user-selected. That creates a locale/language policy concern because the skill appears to force a specific language presentation without explicit user opt-in.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

This skill explicitly tells the agent to fetch a website and query backend endpoints, including switching to an environment that can access the network if needed. The document does not warn users that using the skill will make live external requests, which is a behavior that can affect privacy or system/network integrity under the markdown-file warning criterion.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The output template uses Chinese column names and summary labels as the default format. Without a statement to mirror the user's preferred language, this is a natural-language policy violation under the language/locale rule.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.