Back to skill

Security audit

knowledge-card-factory

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent content-production workflow, but it combines public social posting, cross-platform collection, notifications, checkpointing, and unpinned external skills with inconsistent documentation around approval and target platforms.

Review this skill before installing. Use it only if you are comfortable connecting third-party search, scraping, image-generation, notification, and social-publishing skills. Require a visible preview and explicit approval immediately before every public post, verify the target account and platform, and prefer pinned or reviewed dependency versions where available.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (1)

T08 · Insecure Dependencies

Warning
Location
README.md:180
Finding
Unpinned Third-Party Skill Dependencies## Vulnerability Details **File Location**: `README.md:180-184` **Additional Location**: `workflow.json:16-58` **Vulnerability Type**: Supply-chain risk from mutable third-party dependencies **Risk Level**: Medium ### Vulnerable Code ```bash clawhub install brave-search clawhub install agent-reach clawhub install nano-banana-pro clawhub install card-renderer clawhub install xiaohongshu-mcp ``` These commands install dependencies by name without specifying an exact version, immutable digest, checksum, or verified signature. The corresponding dependency declarations in `workflow.json:16-58` similarly reference mutable marketplace URLs without integrity metadata. ### Technical Analysis The workflow delegates security-sensitive operations to externally maintained Skills, including network searches, cross-platform content collection, image generation, and authenticated social-media publishing. Because dependency versions and artifact identities are not pinned, separate installations can resolve to different dependency implementations. The repository does not vendor or otherwise include the source of these dependencies, so their effective behavior cannot be established from this package alone. If a marketplace account, distribution channel, or dependency release is compromised, an altered dependency could be installed without any change to the audited repository. This is a supply-chain weakness rather than evidence that the currently referenced dependencies are malicious. ### Attack Path 1. An attacker compromises a referenced dependency, its publisher account, or its marketplace distribution entry. 2. The attacker publishes a modified release under the existing dependency name. 3. A user follows the documented unversioned `clawhub install` commands. 4. The installer resolves the dependency name to the attacker-controlled release. 5. The workflow invokes the compromised Skill during search, research, content gen ...[truncated 961 chars]
Remediation
## Remediation Suggestions 1. Pin every dependency to an exact, reviewed version or immutable artifact digest. 2. Add cryptographic checksums or signature verification and fail installation when verification is unavailable or unsuccessful. 3. Maintain a lock file recording resolved versions, artifact hashes, publisher identities, and source locations. 4. Review dependency source code and release changes before updating pinned versions. 5. Restrict each Skill to the minimum required network, filesystem, credential, and account permissions. 6. Isolate research, rendering, and publishing components so content-collection dependencies cannot access publishing credentials. 7. Preserve an explicit confirmation gate immediately before every external publication and enforce it in the publishing integration rather than relying only on workflow configuration. 8. Monitor dependency advisories and marketplace ownership changes, and establish a controlled update and rollback process.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (12)

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The README explicitly describes automatic publication to external social platforms but does not document any explicit user confirmation, review gate, or warning about irreversible posting consequences. In a content-publishing skill, this is dangerous because generated or scraped content could be posted publicly without adequate human review, causing privacy, reputational, compliance, or account-safety issues.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The example workflow shows end-to-end automated posting to Xiaohongshu and returns a published URL, normalizing unattended external posting without highlighting risks or a human approval checkpoint. Because the pipeline also aggregates multi-source content, this raises the chance of publishing inaccurate, copyrighted, policy-violating, or sensitive material directly to a public platform.

Vague Triggers

Medium
Confidence
93% confidence
Finding
The listed activation phrases include broad requests such as "帮我生成内容发小红书" and "自动化内容生产", which could match many ordinary content-assistance requests beyond this specific skill's intended workflow. The documentation does not provide exclusion conditions or constraints to distinguish when this skill should activate versus a simpler drafting or publishing skill.

Missing User Warnings

Medium
Confidence
90% confidence
Finding
The examples show cross-platform scraping and publishing to external services without clearly warning users that data will be transmitted to third parties or that generated content may be posted publicly. In a skill designed for content production and distribution, this omission can cause privacy, compliance, and reputational harm if users do not understand the scope of external access and publication.

Intent-Code Divergence

Medium
Confidence
94% confidence
Finding
The FAQ claims all content requires user confirmation before publication, but multiple examples depict successful publishing as part of the workflow without any visible approval checkpoint. In an automation skill that creates and posts content, this inconsistency is dangerous because users may assume a safety gate exists when it may be skipped, leading to accidental or unauthorized posting.

Intent-Code Divergence

Medium
Confidence
95% confidence
Finding
The documentation presents contradictory statements about supported publishing platforms: the FAQ says only Xiaohongshu is currently supported, while earlier examples show publishing to Weibo, public accounts, and Feishu. This can mislead users and integrators about where content may be sent, increasing the risk of unintended external distribution and weakening informed consent around outbound actions.

Vague Triggers

Medium
Confidence
89% confidence
Finding
The manifest describes the workflow as a 'fully automated solution' from topic selection through publishing, but the activation scope and safety boundaries are not clearly constrained in the description. In a content automation skill that can scrape multiple platforms, generate media, and publish outward, overly broad automation claims can mislead operators about what the system will do autonomously and increase the risk of unintended data collection or publication.

Natural-Language Policy Violations

Medium
Confidence
97% confidence
Finding
The configuration fixes the search language to 'zh', which imposes a specific language/locale behavior. There is no indication that the skill is region-specific by design or that users can opt into or override this locale choice, so it appears to violate the language/locale policy.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
This workflow aggregates content from multiple platforms and sends notifications over external channels, but the manifest does not document what data is collected, retained, forwarded, or shared. In this context, the combination of cross-platform scraping, image extraction, checkpointing, local draft saving, and external notifications raises meaningful privacy and compliance risks, especially if collected content or generated drafts contain personal, copyrighted, or sensitive material.

Natural-Language Policy Violations

Low
Confidence
72% confidence
Finding
The skill documentation is entirely in Chinese and does not indicate that users may choose another language or that the skill is intentionally restricted to a Chinese-language audience. The policy calls for flagging language or locale constraints when they are effectively forced without opt-in or justification.

Natural-Language Policy Violations

Low
Confidence
81% confidence
Finding
The skill is documented for Chinese-language use, but the image-generation prompt template is fixed in English and no user opt-in or language selection is described. This creates a locale/language policy concern because the skill appears to impose a specific language for part of the workflow without explicit user choice or justification.

Natural-Language Policy Violations

Low
Confidence
95% confidence
Finding
A natural-language policy issue exists when a skill forces a specific language without user opt-in. This file presents all instructions and examples only in Chinese and does not indicate that users may choose another language or locale.

Static analysis

No suspicious patterns detected.