Back to skill

Security audit

image-collect

Security checks for vulnerabilities and agentic risk

Overview

The skill processes user-supplied images, performs OCR, and saves the image and extracted text locally as described, with no evidence of deception or destructive behavior.

Install only if you are comfortable with images and OCR text being saved locally. Avoid giving it sensitive images or internal/private URLs unless you intend to keep a local knowledge record, and consider pinning or reviewing dependency versions before production use.

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
  • 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
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (17)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill description understates important behaviors: it accepts remote URLs and data URIs, fetches external content over HTTP, performs OCR, and writes files and database entries locally, yet these capabilities are not clearly declared. This is dangerous because users or orchestrators may invoke the skill expecting simple local image processing, while it actually introduces network exposure, untrusted input handling, and persistent side effects.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The skill description understates important behaviors: it accepts remote URLs and data URIs, fetches external content over HTTP, performs OCR, and writes files and database entries locally, yet these capabilities are not clearly declared. This is dangerous because users or orchestrators may invoke the skill expecting simple local image processing, while it actually introduces network exposure, untrusted input handling, and persistent side effects.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README explicitly documents that the skill can download remote images from URLs and persist both images and extracted knowledge to local storage, but it does not warn users about network access, filesystem writes, or privacy implications. In an agent-skill context, undocumented outbound requests and local data retention increase the risk of SSRF-style misuse, unintended data collection, and silent persistence of sensitive content.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
85% confidence
Finding

The skill advertises executable behavior that can access environment-related capabilities, download inputs, and write local state, but it declares no explicit tool scope or permissions. In an agent setting, missing scope declarations weaken reviewability and can allow the skill to run with broader-than-expected privileges, especially when handling untrusted image URLs and local persistence.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The description omits clear warning that the skill downloads remote images and writes files locally, including updating a JSON database. This lack of transparency is risky because users and higher-level agents may unknowingly permit network access and persistent state changes when supplying an image or URL, increasing the chance of SSRF-style access, unwanted data retention, or filesystem side effects.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The invocation criteria are broad enough that the skill may trigger on ordinary image-related requests, causing unintended downloading, OCR, local file storage, or database modification. In this context, overbroad triggering increases danger because the skill has side effects on both the network and filesystem and may process attacker-controlled image sources.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill’s stated purpose is local image knowledge extraction, but it also accepts arbitrary remote URLs and downloads them over the network. This expands the trust boundary and can enable SSRF-style access to internal resources or unexpected outbound connections, which is risky when not disclosed or constrained by the skill’s declared behavior.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The network fetch capability is broader than necessary for the advertised functionality and allows user-controlled URLs to be retrieved without validation. In skill/agent environments, that can be abused to reach internal services, download maliciously large payloads, or create unintended data flows beyond local-only processing.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The code creates the OCR worker with the fixed language code "chi_sim", which forces Simplified Chinese recognition for all inputs. This is a natural-language/locale policy concern because the skill does not offer user opt-in, language selection, or documented justification for restricting OCR to a specific locale.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The provided manifest context describes a locally focused image knowledge extraction skill, while package.json explicitly describes the package as supporting image download, processing, and OCR and includes axios as a dependency. Downloading images implies network retrieval, which is broader than the stated purpose of extracting knowledge from an image and saving it locally.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The text explicitly says the skill uses Tesseract.js for Chinese OCR recognition, which imposes a language/locale constraint in natural language. Because the file does not present this as an optional user-selected mode or explain a region-specific justification, it matches the language/locale policy concern.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
85% confidence
Finding

Using a caret range for axios allows future compatible releases to be installed implicitly, which can introduce vulnerable or behavior-changing versions without review. In a skill that downloads images over the network, dependency drift in an HTTP client increases supply-chain and request-handling risk.

Content

Scanner excerpt · package.json (reported line 14)May include surrounding context.

json
"README.md"
  ],
  "dependencies": {
    "axios": "^1.6.0",
    "sharp": "^0.33.0",
    "tesseract.js": "^5.0.0"
  },

Unverifiable Dependency: axios has 16 known 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), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
88% confidence
Finding

Axios has multiple known advisories, and because the manifest uses a non-exact version, it is not possible to verify whether the installed release is affected. Since this skill appears to download images, any SSRF, proxy-bypass, credential leakage, or request-manipulation issue in the HTTP client could be reachable in normal operation.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
84% confidence
Finding

Using a caret range for sharp permits unreviewed updates of a complex native/image-processing dependency. Image parsers have a long history of memory-safety and parsing issues, so allowing version drift raises the chance of pulling in a vulnerable release.

Content

Scanner excerpt · package.json (reported line 15)May include surrounding context.

json
],
  "dependencies": {
    "axios": "^1.6.0",
    "sharp": "^0.33.0",
    "tesseract.js": "^5.0.0"
  },
  "license": "MIT"

Unverifiable Dependency: sharp has 4 known advisory(ies) (GHSA-54xq-cgqr-rpm3 (sharp vulnerability in libwebp dependency CVE-2023-4863); GHSA-f88m-g3jw-g9cj (sharp inherited vulnerabilities in libvips: CVE-2026-33327, CVE-2026-33328, CVE-); CVE-2022-29256 (sharp vulnerable to Command Injection in post-installation over build environmen) +1 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
86% confidence
Finding

Sharp and its native image-processing stack have known advisories, and the non-pinned dependency prevents verification of the actual installed version. Because this skill processes image content, any vulnerable parser or native library bug may be exposed directly by attacker-supplied images.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
81% confidence
Finding

Using a caret range for tesseract.js allows dependency changes without explicit review. OCR libraries often process attacker-controlled image input, so unpinned versions can increase exposure to newly introduced security or sandboxing issues.

Content

Scanner excerpt · package.json (reported line 16)May include surrounding context.

json
"dependencies": {
    "axios": "^1.6.0",
    "sharp": "^0.33.0",
    "tesseract.js": "^5.0.0"
  },
  "license": "MIT"
}

Unverifiable Dependency: tesseract.js has 1 known advisory(ies) (GHSA-83rx-c8cr-6j8q (Insecure Default Configuration in tesseract.js)), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
79% confidence
Finding

Tesseract.js has at least one known advisory, and the version range prevents verifying whether the deployed release is affected. In a skill that extracts text from images, insecure defaults or related OCR issues may be directly relevant if untrusted images are processed.

Content

No source excerpt is available for this finding.