Back to skill

Security audit

干饭 skill

Security checks for vulnerabilities and agentic risk

Overview

This food assistant is mostly coherent, but it includes auto-enabled recurring tasks, sensitive location/profile handling, broad map features, and unsafe skill-generation paths that need review before installation.

Review this skill before installing. Use it only if you are comfortable storing a local food profile, sending restaurant/search/routing location data to AMap, and having meal reminders registered by the host. Disable or remove default schedules if you do not want recurring execution, prefer environment variables over saved API-key config files, and avoid generating/installing restaurant skills from untrusted POI or YAML data until slug validation and Markdown/YAML escaping are fixed.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • System PersistenceInstalls backdoors, hooks, services, or scheduled tasks that survive the run
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
Findings (5)

T09 · Insecure Skill Coding Practices

Error
Location
generator/poi-to-skill.mjs:218
Finding

Output Directory Traversal Through Unvalidated Restaurant Slugs

Content
View full analysis
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
generator/poi-to-skill.mjs:108
Finding

Stored Agent Instruction Injection in Generated Skill Files

Content
View full analysis
` - ${k}`).join("\n")} --- # ${r.name} · Skill > 🟡 骨架版 — 基础信息来自高德 API,推荐菜和氛围待社区补充 ## 基本信息 | 项目 | 内容 | |------|------| | 餐厅名称 | ${r.name} | | 品类 | ${r.category} | | 地址 | ${r.address} |`; ``` The standalone generator also inserts unrestricted free-form fields: ```js if (data.special_notes) { md += `\n\n## 特别说明\n\n${data.special_notes}`; } ``` Other directly embedded fields include restaurant names, addresses, categories, directions, Wi-Fi values, atmosphere descriptions, contributor notes, and dish descriptions. ### Technical Analysis The generators produce files intended to be loaded as agent Skill instructions. However, values originating from AMap records, JSON, or YAML are interpolated directly into YAML front matter and Markdown without structural encoding or trust boundaries. An input value containing newlines can: - End or alter YAML front matter. - Add new Skill metadata. - Create new Markdown headings. - Insert imperative instructions directed at the agent. - Override or conflict with the intended restaurant-information behavior. This is more security-sensitive than ordinary Markdown injection because the output is explicitly intended to become agent context or a system-prompt-like Skill. The injected instructions can persist in the generated file and affect future sessions whenever that f ...[truncated 1239 chars]
Remediation
View remediation

T06 · System Persistence

Warning
Location
schedule/eat-schedule.json:1
Finding

Meal Reminder Schedules Enabled for Automatic Persistent Registration

Content
View full analysis
Remediation
View remediation

other

Note
Location
vendor/amap-lbs-skill/SKILL.md:82
Finding

Mandatory AMap Usage Telemetry Before Functional Operations

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
package-lock.json:24
Finding

Unpinned Installer Execution and Third-Party Registry Mirror

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Rogue AgentSelf-Modification, Session Persistence
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (78)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The capability set extends from nearby restaurants to broader travel-style planning and arbitrary interest-point handling. In context, that means the skill can act outside its stated narrow domain, reducing transparency and making over-collection or overreach more likely.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The capability set extends from nearby restaurants to broader travel-style planning and arbitrary interest-point handling. In context, that means the skill can act outside its stated narrow domain, reducing transparency and making over-collection or overreach more likely.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The capability set extends from nearby restaurants to broader travel-style planning and arbitrary interest-point handling. In context, that means the skill can act outside its stated narrow domain, reducing transparency and making over-collection or overreach more likely.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The capability set extends from nearby restaurants to broader travel-style planning and arbitrary interest-point handling. In context, that means the skill can act outside its stated narrow domain, reducing transparency and making over-collection or overreach more likely.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The capability set extends from nearby restaurants to broader travel-style planning and arbitrary interest-point handling. In context, that means the skill can act outside its stated narrow domain, reducing transparency and making over-collection or overreach more likely.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill instructs collection and persistent storage of location, food preferences, budget, and dietary data in user-profile.json without any user-facing privacy notice, retention policy, or consent flow. This is sensitive personal data that can reveal habits, workplace area, and health-related preferences, making the risk materially higher in context.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill omits a warning that nearby-search and navigation requests may transmit user location or location-derived queries to the Gaode/AMap third-party service. Because location data is particularly sensitive, undisclosed sharing undermines informed consent and can expose users to privacy loss beyond the local agent environment.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 140)May include surrounding context.

md
--keywords="烧烤" --city="北京" | node generator/poi-to-skill.mjs --outdir restaurants/

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The discover flow instructs the agent to collect a user location and send it to the AMap API, but the docs do not clearly disclose that this location data is transmitted to a third party. Undisclosed sharing of precise or semi-precise location information is a meaningful privacy vulnerability, especially in an agent setting where users may not realize external services are involved.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The navigation flow sends origin, destination, and travel mode to AMap and may automatically open a browser, yet the documentation lacks a conspicuous warning and explicit consent checkpoint. This combines privacy-sensitive location disclosure with host interaction, making the risk materially higher than a simple content-generation feature.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: axios==1.14.0 — 16 advisory(ies): CVE-2026-44494 (axios Vulnerable to Full Man-in-the-Middle via Prototype Pollution Gadget in `co); CVE-2026-44495 (axios Vulnerable to Credential Theft and Response Hijacking via Prototype Pollut); CVE-2025-62718 (Axios has a NO_PROXY Hostname Normalization Bypass that Leads to SSRF) +13 more

High
Category
Supply Chain
Confidence
94% confidence
Finding

The lockfile pins axios to 1.14.0, and the finding indicates multiple published advisories affecting this version, including SSRF/proxy-bypass and prototype-pollution-related attack paths. In this skill context, a location/restaurant assistant is likely to make outbound HTTP requests to map or nearby-restaurant services, so a vulnerable HTTP client materially increases risk if attacker-controlled URLs, redirects, proxy settings, or response handling are involved.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: form-data==4.0.5 — 1 advisory(ies): CVE-2026-12143 (form-data: CRLF injection in form-data via unescaped multipart field names and f)

High
Category
Supply Chain
Confidence
80% confidence
Finding

form-data 4.0.5 is flagged for CRLF injection via unescaped multipart field names/filenames. If the skill or its vendored dependency ever builds multipart requests using attacker-controlled values, this could enable malformed requests, header injection, or request smuggling behavior against downstream services.

Content

No source excerpt is available for this finding.

YARA rule 'backdoor_persistence': Backdoor persistence with malicious payloads (shell commands, SSH key injection, hidden root users) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · schedule/nudge.mjs (reported line 46)May include surrounding context.

js
} = parseArgs({
  options: {
    meal: { type: "string", short: "m" },
    style: { type: "string", short: "s" },
    json: { type: "boolean", short: "j" },
    help: { type: "boolean", short: "h" },
  },
});

if (args.help) {
  console.log(`
🍜 干饭.skill — 饭点提醒

用法:
  node nudge.mjs [--meal lunch|dinner|late_night] [--style normal|chaotic] [--json]

定时任务示例:
  # crontab
  0 12 * * 1-5  cd /path/to/eat-skill && node schedule/nudge.mjs --meal lunch
  0 18 * * 1-5  cd /path/to/eat-skill && node schedule/nudge.mjs --meal dinner

  # Claude Code
  claude schedule "每天中午12点提醒我吃饭" --command "node schedule/nudge.mjs --meal lunch"
  `);
  process.exit(0);
}

// ═══════════════════════════════════════════
// 趣味文案库
// ═══════════════════════════════════════════

co

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The imported skill advertises broad map, routing, tourism, and heatmap capabilities that substantially exceed the eat-skill’s stated dining-decision purpose. This scope expansion is dangerous because it increases the agent’s reachable behaviors, external network access, and data handling surface without clear justification, making abuse or accidental invocation more likely.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill instructs users to input and locally persist an API key in a config file, and describes runtime key management behavior that goes beyond a simple dining assistant’s expected scope. Local credential storage increases the chance of secret leakage through logs, backups, workspace sharing, or accidental exposure by the agent, especially because the skill is vendored into another skill with a narrower stated purpose.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: axios==1.14.0 — 16 advisory(ies): CVE-2026-44494 (axios Vulnerable to Full Man-in-the-Middle via Prototype Pollution Gadget in `co); CVE-2026-44495 (axios Vulnerable to Credential Theft and Response Hijacking via Prototype Pollut); CVE-2025-62718 (Axios has a NO_PROXY Hostname Normalization Bypass that Leads to SSRF) +13 more

High
Category
Supply Chain
Confidence
97% confidence
Finding

The lockfile pins axios to 1.14.0, and the supplied advisory list includes multiple high-severity issues affecting that version, including SSRF/proxy-bypass and prototype-pollution-related attack chains. In a location/restaurant skill that likely performs outbound HTTP requests to map or nearby-search APIs, a vulnerable HTTP client is especially relevant because attacker-influenced URLs, redirects, proxy settings, or response handling can turn these flaws into credential leakage, SSRF, or request/response manipulation.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: form-data==4.0.5 — 1 advisory(ies): CVE-2026-12143 (form-data: CRLF injection in form-data via unescaped multipart field names and f)

High
Category
Supply Chain
Confidence
86% confidence
Finding

form-data 4.0.5 is flagged for CRLF injection via unescaped multipart field names/filenames. This is a real issue if the skill ever constructs multipart requests using attacker-controlled values, which could let an attacker smuggle or alter headers/body structure when uploading or relaying data to external services; the current skill context makes it somewhat less likely than axios abuse, but still plausible in an integration-heavy agent ecosystem.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: axios==1.14.0 — 16 advisory(ies): CVE-2026-44494 (axios Vulnerable to Full Man-in-the-Middle via Prototype Pollution Gadget in `co); CVE-2026-44495 (axios Vulnerable to Credential Theft and Response Hijacking via Prototype Pollut); CVE-2025-62718 (Axios has a NO_PROXY Hostname Normalization Bypass that Leads to SSRF) +13 more

High
Category
Supply Chain
Confidence
96% confidence
Finding

The allowed axios version range can resolve to axios 1.14.0, which is flagged with multiple advisories including SSRF-related and prototype-pollution-assisted attack paths. Because this skill likely performs outbound HTTP requests to map/location APIs, a vulnerable HTTP client is especially relevant and could enable request manipulation, credential leakage, or response hijacking depending on how the library is used elsewhere.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest describes a food assistant for deciding what to eat and finding nearby restaurants, but this file is explicitly a travel planner that accepts a city, interests like sights/hotels, and route types for multi-stop trip planning. That is a materially different end-user capability than restaurant recommendation or meal choice assistance.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

The README instructs users to run npx clawhub install eat-skill without pinning a specific package version. This can cause users to execute whatever package version is current at install time, increasing supply-chain risk if a malicious or compromised release is published.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The README states that reminders are automatically enabled and also describes collection of user taste, budget, and location data, but does not provide a clear privacy notice or explicit opt-in warning. Users may unknowingly enable background behavior and persist sensitive preference/location information, which creates privacy and consent risks.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The nearby restaurant search and routing features rely on a third-party map service and location-bearing queries, but the README does not explicitly warn users that their search terms and location data may be transmitted to AMap. This omission can lead to unintended disclosure of sensitive location information to an external provider.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README states that scheduled reminders are auto-enabled on install and elsewhere notes the tool learns taste, location, and budget, but it does not present a clear upfront privacy notice or explicit consent flow. Background reminders combined with profile/location use can surprise users and create unintended collection, retention, or exposure of personal data.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
90% confidence
Finding

The README instructs users to run npx clawhub install eat-skill without pinning a version or integrity source. This can cause users to execute whatever package version is current at install time, increasing supply-chain risk if the package is compromised or a breaking/malicious release is published.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · README.md (reported line 167)May include surrounding context.

System Crontab (universal)

bash
crontab -e

# Add these lines (replace with your actual path)
50 11 * * 1-5  cd /path/to/eat-skill && node schedule/nudge.mjs --meal lunch