Back to skill

Security audit

期刊底稿生成器

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious because scanners report substantial capabilities beyond its stated publication-drafting purpose, including remote service discovery and credential transmission, with some analysis incomplete.

The verdict is suspicious, and the scanner recommends not installing. Before considering it, review the remote-service and API-key instructions, confirm any external document processing is acceptable for your materials, and decide whether the extra skill-generation and image functions are within your needs. Scanning was partial: 151 of 155 components were fully inspected, and some analyzer work 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (74)

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · README.en.md (reported line 107)May include surrounding context.

Not required, but strongly recommended — without it nothing errors, you just lose one layer of cross-verification and the semantic layer falls back to manual entry. Three steps to enable: register at https://cuecue.cn → get CUE_API_KEY at https://cuecue.cn/hub/api-key → install cue-research (Omni parsing additionally needs the local Bridge).

The channel route (account-side connectivity self-check; this package ships no connectivity script, so the curl lines below are the whole action):

bash
CATALOG="https://cuecue.cn/api/mcp-catalog"

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · README.en.md (reported line 115)May include surrounding context.

md
rows=json.load(sys.stdin)["data"]["domains"]
[print(r.get("id"), r.get("external_status"), (r.get("routing") or {}).get("url", "-")) for r in rows]'
DOMAIN_ID="<replace with a domain id listed by the line above>"
ROUTING_URL="$(curl -s "$CATALOG" | python3 -c 'import json,sys
rows=json.load(sys.stdin)["data"]["domains"]
hit=[r["routing"]["url"] for r in rows if r.get("id")==sys.argv[1] and r.get("routing")]
print(*hit,sep="\n") if hit else sys.exit("not found: that id is not in the live catalog (or it has no routing.url) - run the listing line first, then fill it in")' "$DOMAIN_ID")"

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · README.md (reported line 141)May include surrounding context.

md
rows=json.load(sys.stdin)["data"]["domains"]
[print(r.get("id"), r.get("external_status"), (r.get("routing") or {}).get("url", "-")) for r in rows]'
DOMAIN_ID="<把这里换成上一行列出的那个域 id>"
ROUTING_URL="$(curl -s "$CATALOG" | python3 -c 'import json,sys
rows=json.load(sys.stdin)["data"]["domains"]
hit=[r["routing"]["url"] for r in rows if r.get("id")==sys.argv[1] and r.get("routing")]
print(*hit,sep="\n") if hit else sys.exit("未取到:该 id 不在目录现值里(或它没有 routing 里的 url 字段)——先跑上面那行列出现场的 id 再填")' "$DOMAIN_ID")"

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · README.md (reported line 133)May include surrounding context.

在 https://cuecue.cn/hub/api-key 取 CUE_API_KEY → 装 cue-research (Omni 解析另需本地 Bridge)。

通道正路(账户侧实连自查;本包不随带自测脚本,下面这几行 curl 就是全部动作):

bash
CATALOG="https://cuecue.cn/api/mcp-catalog"

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · README.md (reported line 133)May include surrounding context.

在 https://cuecue.cn/hub/api-key 取 CUE_API_KEY → 装 cue-research (Omni 解析另需本地 Bridge)。

通道正路(账户侧实连自查;本包不随带自测脚本,下面这几行 curl 就是全部动作):

bash
CATALOG="https://cuecue.cn/api/mcp-catalog"

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied shell snippet implements MCP service discovery and authenticated tool enumeration, not editorial drafting, citation verification, or publication QA. Although the description permits public research, listing remote MCP services and sending a bearer credential to a dynamically selected endpoint are materially different behaviors without an evident connection to that purpose. The snippet does not demonstrate malicious intent or invoke the listed tools, but its actual function is not represented by the declared description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied chunk implements remote MCP discovery and authenticated tool enumeration. It contains no drafting, citation verification, QA checklist generation, or editorial gate logic, and does not establish that the discovered service supports those functions. Although public research could involve network access, arbitrary service discovery and credential-bearing requests are not adequately explained by that declared capability. The snippet does not show execution of remote tools or transmission of manuscript content.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied shell commands perform skill-maintenance checks rather than drafting, citation verification, or editorial QA. The comment indicates that the comparison should identify only derive_skill.py. Although listing files and writing temporary inventories are ordinary implementation details, their purpose here is unrelated to the declared publication workflow. No declared editorial functionality appears in this chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied script's primary operation is skill derivation and packaging, a capability absent from the declared drafting-and-review description. It consumes existing publication specifications rather than issue materials and produces a new skill directory rather than an outline, manuscript, or verified citations. Although the generated instructions describe a related editorial workflow, those instructions do not mean this script performs that workflow: it does not conduct research, verify citations, render a draft, or run the advertised QA gates. Its visible effects are local file reads, directory creation, and file writes/copies; no network access or copied-script execution is shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied script’s primary output is a new skill package, not an outline, manuscript, verified citations, or a completed QA report. Its generated documentation describes drafting and audit workflows, but this script does not execute those workflows. Reading publication specifications and generating reusable assets are related to the declared domain; however, creating an independently installable skill with copied executable scripts is a substantial additional capability absent from the description. No network requests or execution of the copied scripts are shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The code's primary function is diagram creation, not drafting, citation verification, or editorial QA. Although diagrams could support a publication workflow, this is a distinct output capability rather than merely an implementation detail of the declared checks. Its overflow warnings do not implement the promised checks for positioning, word count, reader actions, sentence length, taboo terms, or terminology density. No network access, exfiltration, or unrelated command execution is evident in this chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

Image preparation, prepress checks, and content-reference updates are reasonable supporting details for the declared publication workflow. However, explicitly removing an AI-generation watermark is a distinct, undeclared capability that can remove provenance information; it is not merely a layout or editorial QA check. The script performs local file operations and makes no network requests.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The supplied script's primary output is a new skill package, not the outline, editable manuscript, citation verification, or QA results described. It reads local specifications, extracts reusable publication structure, generates skill instructions, and copies scripts and assets into an output directory. The generated documentation discusses drafting and audit commands, but this script does not execute those workflows. This is a materially different primary capability, not merely an internal rendering detail. No network requests, credential access, or execution of the copied scripts are shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Most image preparation and content patching could reasonably support the declared publication and layout workflow, so those operations alone do not establish a mismatch. However, --crop-watermark explicitly implements removal of an AI-generation label, a distinct capability not represented in the description. It measures image-content boundaries and attempts to remove the relevant bottom or right edge, warning if neither crop is considered safe. The supplied code makes no network requests and shows no credential access or external data transmission.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The supplied chunk primarily reverse-engineers a PDF’s visual design. The description mentions downstream layout checks, but does not disclose reference-PDF design extraction or generation of reusable layout specifications; its tone-only limitation further narrows the stated reference-imitation scope. This is a substantive capability difference, not merely an implementation detail. The script only reads the selected PDF and writes local output files; no network activity, command execution, or exfiltration is evident. The absence of drafting and citation checking in this helper alone is not the basis for the mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The supplied script's primary function is diagram rendering, which is not described in the declared purpose. Although diagrams could support a publication workflow, this is a distinct content-generation capability rather than an implementation detail of the stated editorial checks. The script does not draft manuscripts, retrieve or verify citations, or implement the described editorial QA gates. Its resource access is limited to local specification/font reads and output-file writes; no network access or suspicious execution is evident.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The script's primary checks concern synchronization between a skill generator and example SKILL.md files. They do not evaluate publication layout, word counts, reader actions, sentence length, taboo words, terminology density, or citation provenance. Although this could serve as development tooling for the broader package, its supplied behavior is materially different from the declared editorial workflow and its stated script gates. The visible code shows no network access or exfiltration; the mismatch concerns purpose rather than malicious behavior. The invoked generator's implementation is not supplied, so its additional behavior cannot be assessed.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied command performs diagram generation rather than the described drafting, citation verification, or editorial checks. Although diagrams could support a publication workflow, this chunk shows no connection to those declared functions. The generator's internal behavior is not supplied, so the assessment is limited to the visible invocation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied script's primary purpose is reverse-engineering a sample publication's visual design. This is materially broader than learning its tone and is not simply checking a draft against layout or editorial requirements. The absence of drafting or citation checking in this individual helper is not itself the mismatch; the additional visual-template extraction capability is. The code performs local PDF reads and output-file writes, with no evident network access, exfiltration, or automated editorial sign-off.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

The supplied script's primary function is PDF design analysis and template inference. Although the description mentions downstream layout checks, extracting a sample publication's fonts, palette, geometry, image placements, and page templates is materially different from checking an editorial draft. Its outputs explicitly support visual reproduction, while the description characterizes reference imitation as tone-only. The script performs local PDF reads and output-file writes; no network access, source verification, or editorial sign-off is implemented in this chunk. The mismatch concerns the undeclared design-extraction capability, not merely the absence of other workflow stages.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The supplied chunk primarily generates graphical assets, a capability not disclosed in the description. It does not produce editorial drafts, verify citations, or implement the stated editorial gates. Although diagrams can support publication production, this is a distinct user-facing output capability rather than merely an implementation detail of the declared checks. Its overflow warnings concern diagram geometry, not the described editorial QA. No network access, exfiltration, or command execution is evident.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

Image preparation, prepress checks, and content-JSON updates can reasonably support the declared publication workflow and are not independently treated as mismatches. However, --crop-watermark explicitly invokes cropping intended to remove an AI-generation label, which is a distinct, undeclared capability rather than an editorial or layout QA check. The implementation warns when its margin heuristic finds insufficient room, but otherwise saves the cropped image and can insert its path into publication content.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The script's primary output is a visual production plan, not editorial copy or either of the described validation gates. Although related to publication production, image-slot classification and generation-prompt preparation constitute a substantive capability absent from the description, rather than merely an implementation detail of text drafting or layout checking. The script also adds prompt instructions reserving margins expressly to facilitate downstream cropping of an AI watermark. It does not itself generate images, remove watermarks, invoke rendering scripts, or make network requests.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

The supplied script implements a substantial visual-production planning capability not disclosed by the description's text-drafting, citation-verification, and editorial-QA scope. This goes beyond an incidental implementation detail of the stated layout checks. It does not itself generate images, run a diagram renderer, browse the web, or perform editorial sign-off; its concrete effects are reading local inputs and writing illustration-planning artifacts.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

The visible command is directed at deriving and locally provisioning a publication-specific skill. That is a distinct capability from the declared drafting, citation-checking, and editorial-QA workflow. The script implementation is not supplied, so its exact filesystem effects cannot be verified, and the command does not establish that anything is actually published to a marketplace.

Content

No source excerpt is available for this finding.