Back to skill

Security audit

Image Fetch Toolkit

Security checks for vulnerabilities and agentic risk

Overview

This image-search skill is mostly purpose-aligned, but it asks agents to use many third-party services and optional unpinned package installs without enough scoping, privacy, or supply-chain safeguards.

Install only if you are comfortable with agents making external image and research searches. Avoid using confidential project names, private URLs, unreleased research topics, or personal data in queries unless you explicitly approve sharing them with the selected provider. Do not run the optional npx or uv MCP/image-generation commands unless you pin and review the package version, isolate the environment, and expose only the specific API key required.

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:137
Finding

Unpinned Third-Party Packages May Download and Execute Mutable Code

Content
View full analysis
Remediation
View remediation
uv tool install 'paper-search-mcp==' ``` 2. Use lockfiles and registry-supported integrity hashes where possible. 3. For GitHub-hosted components, reference an immutable reviewed commit SHA rather than a branch or mutable tag. 4. Verify that the package registry publisher corresponds to the linked source repository. 5. Inspect package contents, installation hooks, and transitive dependencies before execution. 6. Require explicit user approval before installing or running any package not already present. 7. Execute third-party tools in an isolated environment with: - A restricted filesystem. - A minimal environment-variable allowlist. - Restricted outbound network access. - No administrative privileges. 8. Pass only the credential required for the selected provider instead of exposing the process to the agent's complete environment. 9. Document a trusted version and update process so package upgrades receive a new security review. ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:69
Finding

API Credentials Are Exposed Through URL Query Parameters

Content
View full analysis
` - **Get key**: https://pixabay.com/api/docs/ - **Rate limit**: 100 requests/min ```bash # Search images curl "https://pixabay.com/api/?key=$PIXABAY_API_KEY&q=mountain+landscape&image_type=photo&per_page=10&safesearch=true" # Search vectors/illustrations curl "https://pixabay.com/api/?key=$PIXABAY_API_KEY&q=robot&image_type=vector&per_page=10" # Search videos curl "https://pixabay.com/api/videos/?key=$PIXABAY_API_KEY&q=ocean&per_page=5" ``` ``` Other credential-bearing URL examples include: ```bash curl "https://api.flickr.com/services/rest/?method=flickr.photos.search&api_key=$FLICKR_API_KEY&text=sunset&license=1,2,4,5,9&per_page=10&format=json&nojsoncallback=1" ``` ```bash curl "https://www.googleapis.com/customsearch/v1?key=$GOOGLE_API_KEY&cx=$GOOGLE_CX&searchType=image&q=AI+conference+2024&num=10&imgSize=large" ``` ```bash curl "https://newsapi.org/v2/everything?q=AI+research&apiKey=$NEWSAPI_KEY&pageSize=10" ``` ```bash curl "https://gnews.io/api/v4/search?q=climate+change&token=$GNEWS_API_KEY&lang=en&max=10" ``` ### Technical Analysis The examples interpolate API credentials directly into URL query strings passed as command-line arguments. Although the requests target services relevant to the declared image-search functionality, putting secrets in URLs increases their exposure beyond the minimum required for authentication. Expanded command-line URLs may be recorded or exposed through: - Shell history. - Process inspection tools while `curl` is running. - Terminal session capture and command tracing. - Agent execution logs ...[truncated 1877 chars]
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (22)

Vague Triggers

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill description says to use the skill for "any purpose" and across many common content tasks, which can cause over-invocation on broad user prompts. Overly broad activation increases the chance that user queries, URLs, or sensitive research topics are unnecessarily routed to networked workflows and third-party services.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 384)May include surrounding context.

md
### Instagram (via Meta Graph API)
- **Base URL**: `https://graph.instagram.com`
- **Auth**: Access Token
- **Get key**: https://developers.facebook.com/docs/instagram-api

---

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The quick-start section presents a general-purpose image workflow and enumerates multiple external tools before defining when the skill should be activated. This can encourage agents to default to third-party network actions without first checking necessity, consent, or safer local alternatives.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill extensively documents calls to external image, academic, news, e-commerce, and social APIs, but it does not warn that prompts, search terms, URLs, or fetched targets may be transmitted to third parties. In agent contexts, this omission can lead to unintentional disclosure of sensitive user intent, private URLs, or research topics.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
87% confidence
Finding

This command sends search queries to the Unsplash API along with authentication material in the request header, causing user-supplied topics to leave the local environment. The danger is contextual rather than inherently malicious: if the query contains confidential project names, internal topics, or personal data, the skill facilitates third-party disclosure without warning.

Content

Scanner excerpt · SKILL.md (reported line 34)May include surrounding context.

bash
# Search photos
curl -H "Authorization: Bearer $UNSPLASH_ACCESS_KEY" \
  "https://api.unsplash.com/search/photos?query=renewable+energy&per_page=10&orientation=landscape"

# Get a random photo

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 35)May include surrounding context.

bash
# Search photos
curl -H "Authorization: Bearer $UNSPLASH_ACCESS_KEY" \
  "https://api.unsplash.com/search/photos?query=renewable+energy&per_page=10&orientation=landscape"

# Get a random photo
curl -H "Authorization: Bearer $UNSPLASH_ACCESS_KEY" \

External Transmission

Medium
Category
Data Exfiltration
Confidence
82% confidence
Finding

The documented random-photo request transmits query terms to Unsplash and may trigger external retrieval based on user input. While normal for an image-fetch skill, it still creates data-sharing risk if agents use it for sensitive or proprietary topics without notice.

Content

Scanner excerpt · SKILL.md (reported line 39)May include surrounding context.

md
# Get a random photo
curl -H "Authorization: Bearer $UNSPLASH_ACCESS_KEY" \
  "https://api.unsplash.com/photos/random?query=technology&orientation=squarish"

# Download tracked photo (respects photographer attribution)
curl -H "Authorization: Bearer $UNSPLASH_ACCESS_KEY" \

External Transmission

Medium
Category
Data Exfiltration
Confidence
80% confidence
Finding

The photo download instruction causes outbound access to a third-party endpoint and retrieval of remote content, which can expose access patterns and transfer potentially unreviewed media into the environment. In an agent setting, repeated automated downloads also increase tracking and content-safety exposure.

Content

Scanner excerpt · SKILL.md (reported line 43)May include surrounding context.

Download tracked photo (respects photographer attribution)

curl -H "Authorization: Bearer $UNSPLASH_ACCESS_KEY"
-L "https://api.unsplash.com/photos/<PHOTO_ID>/download"

text

**Response fields**: `results[].urls.raw`, `results[].urls.full`, `results[].urls.regular`, `results[].urls.small`, `results[].user.name`, `results[].links.html`

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 49)May include surrounding context.

md
**Response fields**: `results[].urls.raw`, `results[].urls.full`, `results[].urls.regular`, `results[].urls.small`, `results[].user.name`, `results[].links.html`

### Pexels API
- **Base URL**: `https://api.pexels.com/v1`
- **Auth**: Header `Authorization: <API_KEY>`
- **Get key**: https://www.pexels.com/api/
- **Rate limit**: 200 requests/hour

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

This search example sends user-derived queries to Pexels, which can disclose sensitive topics or identifiers to a third-party service. The skill context makes this expected behavior, but the absence of privacy warnings or activation controls increases the operational risk.

Content

Scanner excerpt · SKILL.md (reported line 57)May include surrounding context.

bash
# Search photos
curl -H "Authorization: $PEXELS_API_KEY" \
  "https://api.pexels.com/v1/search?query=nature&per_page=10&orientation=landscape"

# Curated photos
curl -H "Authorization: $PEXELS_API_KEY" \

External Transmission

Medium
Category
Data Exfiltration
Confidence
76% confidence
Finding

The curated-photos request contacts a third-party service and reveals use of the skill and associated credentials, though it does not inherently include user-provided sensitive content. Its risk is lower than query-bearing requests, but it still represents undocumented external transmission.

Content

Scanner excerpt · SKILL.md (reported line 61)May include surrounding context.

md
# Curated photos
curl -H "Authorization: $PEXELS_API_KEY" \
  "https://api.pexels.com/v1/curated?per_page=15"

# Get photo by ID
curl -H "Authorization: $PEXELS_API_KEY" \

External Transmission

Medium
Category
Data Exfiltration
Confidence
78% confidence
Finding

Fetching a photo by ID from Pexels is an external network action that may reveal usage metadata and imports remote content into the workflow. This is routine behavior for the skill, but still should be disclosed and constrained to approved cases.

Content

Scanner excerpt · SKILL.md (reported line 65)May include surrounding context.

Get photo by ID

curl -H "Authorization: $PEXELS_API_KEY"
"https://api.pexels.com/v1/photos/<PHOTO_ID>"

text

### Pixabay API

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 88)May include surrounding context.

md
**Categories**: `backgrounds`, `fashion`, `nature`, `science`, `education`, `feelings`, `health`, `people`, `religion`, `places`, `animals`, `industry`, `computer`, `food`, `sports`, `transportation`, `travel`, `buildings`, `business`, `music`

### Flickr API
- **Base URL**: `https://api.flickr.com/services/rest`
- **Auth**: API key query param
- **Get key**: https://www.flickr.com/services/apps/
- **Note**: Check license! Use `license=1,2,4,5,9` for CC-compatible

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

The Flickr search request transmits query text and an API key to an external service, which can leak sensitive interests or project details if used incautiously. The line also encourages use of third-party content retrieval without a corresponding privacy and consent warning.

Content

Scanner excerpt · SKILL.md (reported line 95)May include surrounding context.

bash
# Search photos (CC license only)
curl "https://api.flickr.com/services/rest/?method=flickr.photos.search&api_key=$FLICKR_API_KEY&text=sunset&license=1,2,4,5,9&per_page=10&format=json&nojsoncallback=1"

# Get photo URL: https://farm{farm}.staticflickr.com/{server}/{id}_{secret}.jpg

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 116)May include surrounding context.

text

### Bing Image Search API
- **Base URL**: `https://api.bing.microsoft.com/v7.0/images/search`
- **Auth**: Header `Ocp-Apim-Subscription-Key: <KEY>`
- **Get key**: https://www.microsoft.com/en-us/bing/apis/bing-image-search-api
- **Rate limit**: 1000 transactions/month (free)

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

This Bing search example sends query contents to a third-party image-search provider and uses a subscription key in the request, creating normal but meaningful data-transmission risk. If agents invoke it broadly, sensitive user topics may be exposed externally.

Content

Scanner excerpt · SKILL.md (reported line 123)May include surrounding context.

bash
curl -H "Ocp-Apim-Subscription-Key: $BING_API_KEY" \
  "https://api.bing.microsoft.com/v7.0/images/search?q=cute+cats&count=10&imageType=Photo&size=Large"

Tavily Extract (for webpage images)

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The skill instructs agents to install and run an external MCP-related package via npx skills add openags/paper-search-mcp without pinning a specific version or commit. This creates a supply-chain risk: future package updates or dependency compromise could cause different, potentially malicious code to be fetched and executed in agent environments.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
84% confidence
Finding

The Semantic Scholar request transmits academic search queries to an external API, which may reveal confidential research focus or unpublished work. In an academic-image skill, this is contextually expected, but the lack of caution about sensitive research topics makes it risky.

Content

Scanner excerpt · SKILL.md (reported line 149)May include surrounding context.

bash
# Search papers
curl "https://api.semanticscholar.org/graph/v1/paper/search?query=protein+folding&limit=10&fields=title,url,openAccessPdf"

Semantic Scholar API (Free)

External Transmission

Medium
Category
Data Exfiltration
Confidence
84% confidence
Finding

This example sends paper-related queries, including figure-oriented retrieval, to Semantic Scholar and may expose sensitive scientific interests or undisclosed internal work. The skill makes such lookups easy but does not instruct agents to assess sensitivity first.

Content

Scanner excerpt · SKILL.md (reported line 155)May include surrounding context.

Semantic Scholar API (Free)

bash
# Search papers with figure references
curl "https://api.semanticscholar.org/graph/v1/paper/search?query=transformer+architecture&limit=5&fields=title,url,openAccessPdf,figures"

arXiv API (Free, no key needed)

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% 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 · SKILL.md (reported line 218)May include surrounding context.

md
max_w = max(img.width for img in images)
    max_h = max(img.height for img in images)
    
    # Create canvas
    total_w = cols * max_w + (cols + 1) * padding
    total_h = rows * max_h + (rows + 1) * padding
    canvas = Image.new('RGB', (total_w, total_h), bg_color)

Rp1

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The command npx mcp-image invokes a remotely resolved package without a pinned version, allowing unreviewed code changes to be pulled at execution time. In an agent skill, this is especially risky because operators may treat the documentation as safe-by-default and execute it in privileged environments.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest description says the skill can 'Search and fetch images from the internet' across many broad categories, but it does not define specific trigger phrases, scope limits, or exclusion conditions. In a manifest file, this kind of open-ended wording can overlap with common user requests for images and make invocation boundaries unclear.

Content

No source excerpt is available for this finding.