Back to skill

Security audit

Startup Toolkit

Security checks for vulnerabilities and agentic risk

Overview

This skill is a simple startup-project scaffold with some overbroad claims and a path-handling weakness, but no hidden access, persistence, credential use, network calls, or destructive behavior was found.

Install only if you want a very minimal scaffold helper. Treat the auth, payments, analytics, and CI/CD claims as aspirational unless separately verified, and pass a simple project name such as my-startup rather than paths, spaces, wildcards, or option-like values.

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
launch.sh:4
Finding
Unvalidated and Unquoted Project Path Input## Vulnerability Details **File Location**: `launch.sh`, lines 4–9 **Vulnerability Type**: Unvalidated path input, shell word splitting, and pathname expansion **Risk Level**: Medium ### Vulnerable Code ```bash NAME="${1:-mystartup}" echo "🚀 Launching $NAME..." echo "📦 Creating project structure..." mkdir -p $NAME/{frontend,backend,database,docker} ``` ### Technical Analysis The first command-line argument is accepted as a project path without validation and subsequently expanded without quotation in the `mkdir` command. Because `$NAME` is unquoted, Bash applies word splitting and pathname expansion to its value. Input containing whitespace or wildcard characters can therefore produce multiple path arguments. The script also permits absolute paths and `..` traversal components, allowing directory creation outside the intended working directory. The command does not use the conventional `--` end-of-options separator. Option-like input may consequently be interpreted by `mkdir` rather than strictly as a path, depending on the resulting expansion. This is not direct arbitrary command execution: shell metacharacters introduced through ordinary parameter expansion are not reparsed as shell syntax. The confirmed issue is uncontrolled filesystem path selection and expansion. ### Attack Path 1. An attacker or untrusted automation source controls the first argument supplied to `launch.sh`. 2. The attacker supplies an absolute path, traversal sequence, whitespace-containing value, or glob pattern. 3. The script assigns that value to `NAME` without checking its syntax or destination. 4. Bash performs word splitting and pathname expansion when evaluating the unquoted `$NAME`. 5. `mkdir -p` creates `frontend`, `backend`, `database`, and `docker` directory trees at one or more unintended locations where the invoking user has write permission. For example, a traversal-based argument can direct creation outside the curre ...[truncated 736 chars]
Remediation
## Remediation Suggestions 1. Validate the project name against a strict allowlist and reject empty values, absolute paths, path separators, and traversal components. 2. Quote all variable expansions used as filesystem paths. 3. Add `--` before path operands to terminate option processing. 4. Resolve and verify the destination against an explicitly permitted base directory when callers are allowed to provide nested paths. 5. Exit with a clear error when validation fails. Example hardened implementation: ```bash #!/bin/bash set -euo pipefail NAME="${1:-mystartup}" if [[ ! "$NAME" =~ ^[A-Za-z0-9._-]+$ ]] || [[ "$NAME" == "." || "$NAME" == ".." ]]; then printf 'Error: invalid project name\n' >&2 exit 1 fi echo "🚀 Launching $NAME..." echo "📦 Creating project structure..." mkdir -p -- "$NAME"/{frontend,backend,database,docker} ``` If nested destinations are a required feature, canonicalize the destination and verify that it remains beneath a trusted base directory before creating anything.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

Vague Triggers

Medium
Confidence
87% confidence
Finding
The description claims the skill provides 'everything you need to launch a startup' across multiple sensitive domains including landing pages, authentication, payments, and analytics. This is overly broad because it invites activation for many loosely related tasks and can cause the agent to over-trust or over-apply the skill in contexts involving sensitive operations such as auth or payments.

Static analysis

No suspicious patterns detected.