Back to skill

Security audit

Imagen Search

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious with medium confidence because it reports image-search features alongside undeclared analytics and image-request submission, and the scan covered only part of the artifact.

Consider the undeclared analytics and image-request submission, including the reported transmission of contact details, before installing. The verdict is suspicious, not a claim of malicious intent. Scanning was partial: 7 of 13 components were fully inspected, 6 partly inspected, and tool-misuse analysis was degraded, so the evidence is incomplete.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (11)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

The supplied commands concern analytics rather than image discovery or retrieval. Keyword analytics is a distinct data-handling capability not described by the declared purpose. The explicit-consent comment does not eliminate that disclosure mismatch. Because the underlying script is not supplied, the snippet does not establish what data is collected, stored, or transmitted.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
84% confidence
Finding

The declared purpose covers image search and retrieval, but the supplied command invokes a distinct demand-submission operation. That capability is not described. The CLI implementation is absent, so the snippet does not establish whether submission generates images, publishes catalog entries, or performs other downstream actions.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read, file_write, network.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 22)May include surrounding context.

md
e requested search: continue using the installed version and mention the check failure briefly. Do not retry each search, run background polling or install scheduled tasks.

If `update_available` is true, show the installed/latest versions and relevant changes, then offer an upgrade. Continue the original task if the user defers; do not repeatedly ask in that conversation. Never replace installed files without upgrade authorization. If the user has already requested an upgrade, carry it through without asking the same permission again. Follow https://imagen.chat/install/imagen-search.md for an authorized update, compare versions and changes, verify the ZIP SHA256, and keep a backup. Credentials remain outside the Skill in the existing local configuration; do not print or move them. Preserve project manifests and user files. Modified package files must not be silently overwritten. Refresh the client after an update, verify the installed version and run a free search. Roll back the retai

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 52)May include surrounding context.

node /scripts/imagen.mjs download --id --original --max-credits 1 --operation-id --out /original.png

text

Use the user's **Imagen creator API key**, never a model-provider key, publisher token or browser cookie. The optional local page guides users to `https://imagen.chat/creator`: sign in, create a named key, allow unlocking if desired, and paste the secret locally. Never request keys in chat, expose them in output, or pass them in command arguments or image URLs. Keys are saved owner-readable only in `~/.config/imagen-chat/creator.json`; never read this credential file into model context or distribute it. Local removal does not revoke the remote key. Server keys expire after 90 days and can be revoked on Imagen. Configuration and validation do not spend credits.

Both original operations require an authenticated account and reuse the same entitlement flow:

The skill can submit catalog-content requests and contact information, a capability outside its stated image-search and retrieval purpose.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

submitDemand posts a user-provided email, image request, and locale to /gallery/demand-requests, creating a remote catalog suggestion. This is a separate content-request workflow, not an implementation detail of searching, downloading images, or account setup. Preview and confirmation checks constrain submission but do not make this additional capability part of the declared scope.

Content

No source excerpt is available for this finding.

User-facing Markdown and notices are hardcoded in Chinese without a language choice.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

Image labels, attribution, the watermark notice, and the no-results response are emitted in Chinese regardless of the user's language. These output functions provide no locale selection or opt-in, and the file gives no justification for restricting these messages to Chinese.

Content

No source excerpt is available for this finding.

The write-serialization comment contradicts the uncoordinated read–modify–write implementation.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

Concurrent calls can read the same manifest and each write a replacement based on that stale state, causing the last rename to silently discard another call's placements or revision history. Unique temporary files and atomic rename prevent partial writes but do not serialize the read–modify–write transaction; the effect is limited to manifest integrity and does not establish code execution or credential exposure.

Content

No source excerpt is available for this finding.

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · LICENSE (reported line 7)May include surrounding context.

text
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

The skill mandates Chinese image labels without user language choice or opt-in.

Low
Category
Output Handling
Confidence
94% confidence
Finding

The instruction to keep the visible “图 1 / 图 2” labels requires Chinese text in user-facing output regardless of the user's language. Unlike the watermarked-result notice, which explicitly uses the user's language, this requirement provides no localization option or justification.

Content

No source excerpt is available for this finding.

The generated preview uses fixed Chinese-language text without a user language choice or documented locale justification.

Low
Category
Output Handling
Confidence
90% confidence
Finding

The preview heading, guidance, and suggested follow-up are hardcoded in Chinese regardless of the user's language. The function provides no locale selection or opt-in, so generated user-facing content imposes a specific language.

Content

No source excerpt is available for this finding.