Back to skill

Security audit

fliggy-trip-planner

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent travel planner, but it tells agents to automatically install an unpinned global npm CLI and run shell-style commands with user-derived travel inputs.

Review before installing. Only use this skill in an environment where installing and running the flyai CLI is acceptable, and require explicit approval or preinstallation of a reviewed flyai version. Treat generated prices, schedules, images, and booking links as unverified travel data that should be checked with official or booking sources before purchase.

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:21
Finding

Unpinned Global Installation of a Third-Party CLI Package

Content
View full analysis
/dev/null 2>&1 || npm i -g @fly-ai/flyai-cli ``` The surrounding instructions require this installation to occur before any trip-planning operation and direct the agent to execute it immediately if `flyai` is unavailable. ### Technical Analysis The skill installs `@fly-ai/flyai-cli` globally without specifying a reviewed version, validating a cryptographic integrity hash, using a lockfile, or obtaining explicit user approval. Consequently, the package version and effective installation payload can change after the skill itself has been audited. An npm installation can execute package lifecycle scripts such as `preinstall`, `install`, and `postinstall`. Those scripts run with the privileges of the account executing the agent. The global `-g` installation also modifies the user's shared environment rather than an isolated project directory. Although the repository does not establish that the named package is currently malicious, the mandatory mutable installation creates a supply-chain exposure. A compromised maintainer account, malicious future release, registry compromise, or dependency-level compromise could turn an otherwise ordinary trip-planning request into arbitrary local code execution. ### Attack Path 1. An attacker compromises the npm package, one of its transitive dependencies, or the associated publishing account. 2. The attacker publishes a malicious release or modifies an installation-time dependency. 3. A user invokes this skill for a trip-planning request on a system where `flyai` is not already installed. 4. The skill instructs the agent to execute `npm i -g @fly-ai/flyai-cli`. 5. npm retrieves the mutable package release and executes any applicable lifecycle scripts. 6. Malicious code runs with the privileges of ...[truncated 785 chars]
Remediation
View remediation
``` 2. Use a project-local dependency and a committed lockfile rather than a global installation. 3. Enforce package integrity through a lockfile, verified registry metadata, and approved cryptographic hashes where the package-management workflow supports them. 4. Review the selected package version, its transitive dependencies, and all npm lifecycle scripts before approval. 5. Require explicit user consent before downloading or installing software. 6. Install the CLI in a restricted sandbox or container with minimal filesystem, credential, and network access. 7. Consider disabling lifecycle scripts with `--ignore-scripts` if the package functions correctly without them. 8. Fail safely when the reviewed dependency is unavailable instead of automatically installing the newest registry release. ]]>

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:47
Finding

Command Injection Risk Through Unquoted User-Derived CLI Arguments

Content
View full analysis
--destination <目的地> --dep-date <日期> --back-date <返程日期> --sort-type 3 Step 2: 搜索酒店 flyai search-hotel --dest-name <目的地> --poi-name <核心景点> --check-in-date <入住> --check-out-date <退房> --sort rate_desc --max-price <预算上限> Step 3: 搜索景点 flyai search-poi --city-name <目的地> --poi-level 5 flyai keyword-search --query "<目的地> 美食/特色体验" ``` ### Technical Analysis The documented command templates place values extracted from user input directly into shell commands. Most placeholders are unquoted, and the one quoted query does not define validation or escaping requirements. The skill also does not require structured process invocation, strict schemas, allowlists, or an end-of-options delimiter. If an agent materializes these templates as shell command strings, shell metacharacters in a destination, origin, point-of-interest name, date, or budget value can be interpreted by the shell rather than passed as literal data. Relevant syntax can include command separators, substitutions, redirections, pipes, and newline characters. Even where shell metacharacters are escaped, values beginning with `-` may be interpreted as additional CLI options unless the invoked command strictly binds each value or supports an end-of-options delimiter. The exact consequences of option injection depend on the behavior supported by `flyai`. ### Attack Path 1. An attacker supplies a crafted trip-planning parameter, such as a destination or point-of-interest name containing shell syntax or a value designed to introduce an unintended CLI option. 2. The skill extracts the parameter and inserts it into one of the documented command templates. 3. The agent executes the resulting text through a command shell without strict validation and argument-safe ...[truncated 1174 chars]
Remediation
View remediation
Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (6)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill directs the agent to immediately run a global npm installation when a dependency is absent, with no user-facing warning, consent, or safety check. This is dangerous because it authorizes remote code acquisition and execution from the package ecosystem as part of routine handling, creating supply-chain and host-integrity risk disproportionate to the task of itinerary planning.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The description uses broad activation language covering general trip planning and vague queries like where to travel, which can cause the skill to trigger in situations the user did not clearly intend. Overbroad activation increases the chance that the agent will run the skill's side-effecting workflow, including tool use and dependency checks, on loosely related requests.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The trigger scenarios are ambiguous and broad, allowing activation from common phrases such as asking where to go or requesting vacation ideas. In this skill, ambiguous triggering is more dangerous because activation leads into a workflow that may invoke external tools and even attempt package installation, so accidental invocation has real operational impact.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill explicitly instructs the agent to install a global npm package if a CLI is missing, which expands the agent's capabilities by permitting package installation and arbitrary third-party code execution on the host. In the context of a travel-planning skill, automatic environment modification is unnecessary and dangerous because it can introduce supply-chain risk, persistence, and unintended system changes without informed user approval.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README presents flight, hotel, attraction, and budget outputs as authoritative examples without warning users that prices, availability, routes, and opening information can change quickly. In a travel-planning skill, stale or inaccurate information can mislead users into making poor booking or scheduling decisions, especially when they rely on the generated itinerary without independent verification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
74% confidence
Finding

The skill content is entirely written in Chinese and the examples, trigger phrases, and output expectations implicitly assume Chinese-language operation, without offering a language choice or documenting a justified locale restriction. This can violate language/locale policy when the user has not opted into Chinese or a China-specific workflow.

Content

No source excerpt is available for this finding.