Back to skill

Security audit

research-to-wechat

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent WeChat article-production skill, but it deserves Review because installation and runtime can perform broad networked publishing actions with credentials and imperfect scoping.

Install through ClawHub rather than the curl-to-bash manual command. Before using draft save, image upload, rewrite review, browser-assisted delivery, or multi-platform distribution, confirm exactly which article text, source material, images, credentials, and platforms will be used. Review or disable EXTEND.md/CTA behavior if you do not want promotional blocks in drafts.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (59)

Chaining Abuse

High
Category
Tool Misuse
Confidence
99% confidence
Finding

Piping a network response directly into bash is a well-known unsafe pattern because it combines remote retrieval and execution into a single unreviewable step. In the context of an install command, this materially increases the chance of supply-chain compromise or accidental execution of tampered content.

Content

Scanner excerpt · README.md (reported line 15)May include surrounding context.

Manual:

bash
curl -fsSL https://raw.githubusercontent.com/Fei2-Labs/skill-genie/main/research-to-wechat/scripts/install-openclaw.sh | bash

Optional runtime integrations:

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The documented purpose omits materially relevant behaviors such as external fetching and use of a spoofed WeChat user agent to retrieve content. Hidden or downplayed network collection behavior is dangerous because users may provide sensitive inputs assuming a local writing pipeline, while the skill actually reaches out to third-party sites and may bypass access controls or anti-bot expectations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The documented purpose omits materially relevant behaviors such as external fetching and use of a spoofed WeChat user agent to retrieve content. Hidden or downplayed network collection behavior is dangerous because users may provide sensitive inputs assuming a local writing pipeline, while the skill actually reaches out to third-party sites and may bypass access controls or anti-bot expectations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The documented purpose omits materially relevant behaviors such as external fetching and use of a spoofed WeChat user agent to retrieve content. Hidden or downplayed network collection behavior is dangerous because users may provide sensitive inputs assuming a local writing pipeline, while the skill actually reaches out to third-party sites and may bypass access controls or anti-bot expectations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The documented purpose omits materially relevant behaviors such as external fetching and use of a spoofed WeChat user agent to retrieve content. Hidden or downplayed network collection behavior is dangerous because users may provide sensitive inputs assuming a local writing pipeline, while the skill actually reaches out to third-party sites and may bypass access controls or anti-bot expectations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The documented purpose omits materially relevant behaviors such as external fetching and use of a spoofed WeChat user agent to retrieve content. Hidden or downplayed network collection behavior is dangerous because users may provide sensitive inputs assuming a local writing pipeline, while the skill actually reaches out to third-party sites and may bypass access controls or anti-bot expectations.

Content

No source excerpt is available for this finding.

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
97% confidence
Finding

The anti-refusal language uses mandatory phrasing like 'must execute' and tells the agent not to skip steps because of context decay, which pressures the model to follow embedded instructions rigidly. In an adversarial or mis-scoped environment, this can override safer host-level judgment, compel unnecessary shell/file operations, and make prompt-injection resistance weaker by framing compliance with the skill text as mandatory.

Content

Scanner excerpt · SKILL.md (reported line 251)May include surrounding context.

md
`draft/update` returning `errcode:0` does **not** mean the content landed. Observed 2026-09-25: the API reported ok while the live draft still held the previous title and body. Never judge success from the write response, and never report success to the user from it.

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
80% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · scripts/_wechat_design_catalog.py (reported line 133)May include surrounding context.

python
def choose_text(texts: list[dict[str, object]], include: tuple[str, ...], exclude: tuple[str, ...], size: int) -> dict[str, object]:
    for node in texts:
        name = str(node.get("name", "")).lower()
        if any(token in name for token in include) and not any(token in name for token in exclude):
            return node
    ranked = sorted(texts, key=lambda item: abs(int(item.get("fontSize", size) or size) - size))
    return ranked[0] if ranked else {}

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
80% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · scripts/_wechat_design_catalog.py (reported line 141)May include surrounding context.

python
def choose_text(texts: list[dict[str, object]], include: tuple[str, ...], exclude: tuple[str, ...], size: int) -> dict[str, object]:
    for node in texts:
        name = str(node.get("name", "")).lower()
        if any(token in name for token in include) and not any(token in name for token in exclude):
            return node
    ranked = sorted(texts, key=lambda item: abs(int(item.get("fontSize", size) or size) - size))
    return ranked[0] if ranked else {}

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
80% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · scripts/_wechat_design_catalog.py (reported line 147)May include surrounding context.

python
def choose_text(texts: list[dict[str, object]], include: tuple[str, ...], exclude: tuple[str, ...], size: int) -> dict[str, object]:
    for node in texts:
        name = str(node.get("name", "")).lower()
        if any(token in name for token in include) and not any(token in name for token in exclude):
            return node
    ranked = sorted(texts, key=lambda item: abs(int(item.get("fontSize", size) or size) - size))
    return ranked[0] if ranked else {}

Chaining Abuse

High
Category
Tool Misuse
Confidence
97% confidence
Finding

The explicit '| bash' pattern creates an immediate code-execution chain from untrusted network content to a shell interpreter. In an installer context this is especially dangerous because users are primed to trust and run it, potentially granting the fetched script full access to the user's files, tokens, and configuration.

Content

Scanner excerpt · scripts/install-openclaw.sh (reported line 9)May include surrounding context.

sh
# For ClawHub users: clawhub install research-to-wechat
#
# Usage:
#   curl -fsSL https://raw.githubusercontent.com/Fei2-Labs/skill-genie/main/research-to-wechat/scripts/install-openclaw.sh | bash
#

set -e

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · scripts/jev_rewrite.py (reported line 928)May include surrounding context.

python
parser = build_parser()
    args = parser.parse_args(list(argv) if argv is not None else None)
    out = stdout if stdout is not None else sys.stdout
    environment = env if env is not None else dict(os.environ)

    if args.policy:
        emit(out, policy())

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 · CHANGELOG.md (reported line 35)May include surrounding context.

md
### Added

- `references/original-rewrite.md` — canonical contract for the optional `rewriteMode: methodology-only`: activation (readable source material AND explicit rewrite intent, both required), the intake evidence record, positive/negative trigger examples, private source separation, methodology brief and exclusion ledger, the independent research ledger, persona resolution, the five Jev evaluation dimensions, gate policy `screening-policy-1`, round records, `rewrite-report.md`, and version binding after render normalization
- `scripts/jev_rewrite.py` — standard-library-only screening evaluator with offline `--help`, `--policy`, `--example-input`, `--example-output`; credential comes from `TYPESAFE_API_KEY` in the environment
- `tests/` — offline unit and documentation/regression tests; no network call and no WeChat write

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README documents fetching external URLs and uploading rendered content and images to WeChat, but it does not clearly warn that source material, article text, metadata, and images may be transmitted to third-party services. In a research and publishing pipeline, that omission can lead users to unknowingly send sensitive or copyrighted material off-system, creating privacy, confidentiality, and compliance risk.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
95% confidence
Finding

The skill advertises shell, network, file read/write, and environment-dependent behavior but does not declare an explicit tool scope or permissions boundary. That makes it easier for a host agent to invoke it with broader capabilities than a reviewer or user would reasonably expect, increasing the chance of unintended data access, network exfiltration, or unsafe command execution.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The invocation text is extremely broad, covering generic writing, rewriting, formatting, HTML conversion, draft generation, and distribution. Overbroad routing language can cause an agent to trigger this high-capability skill for ordinary writing tasks, unnecessarily granting shell/network/file access and increasing the blast radius if the skill mishandles data or performs remote fetches.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Much of the asset is dedicated to paid-group/community CTAs, pricing blocks, QR placeholders, and subscription-style promotions rather than research conversion features. In a publishing/distribution skill, this creates a real risk that generated articles or exported WeChat drafts will include unrelated monetization prompts, causing deceptive output, policy violations, brand damage, or unauthorized promotion.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The design repeatedly hard-codes user-facing copy in Chinese, starting with labels such as "付费社群" and continuing throughout the file. Because this file provides no language option, opt-in, or region-specific justification, it violates the language/locale policy constraint for natural-language content.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This design asset contains a large library of themed article and mockup templates that are unrelated to the stated research-to-WeChat pipeline. In an agent skill, mismatched bundled assets can steer downstream generation toward unintended outputs, increase prompt/context confusion, and create a covert channel for off-purpose behavior such as injecting promotional or misleading content into published drafts.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The example explicitly instructs saving generated article content to a WeChat draft box but does not tell the agent to obtain or confirm user consent for transmitting potentially sensitive source material to an external third-party platform. In a research-to-publishing skill, inputs may contain proprietary notes, transcripts, or unpublished writing, so silent draft upload creates a meaningful data handling and privacy risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The multi-platform distribution example directs the system to repurpose content for Xiaohongshu, Jike, and Moments without any warning that the user's material will be transformed and prepared for additional external platforms. This increases the risk of unintended disclosure, broader dissemination than the user realized, and policy/privacy issues if the original source material was only meant for WeChat or local drafting.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The sample EXTEND.md content embeds Chinese-only headings, body text, and engagement prompts as the apparent default author configuration. Because the document does not state that this is optional locale-specific content or offer language selection, it can be read as prescribing a fixed language, which conflicts with language/locale neutrality expectations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The guide tells users to run update-cover as the preferred method before immediately presenting the command, but the critical side effect—that any draft/update clears editor-only fields like 原创、赞赏、合集—is documented only in a later section. In an operational publishing workflow, separating the destructive warning from the command increases the chance a user will follow the example and unintentionally wipe production draft metadata.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The design catalog presents all template names in Chinese (e.g. 极简, 编辑, 杂志) and the workflow depends on those labels for selection, but the document does not state that the skill is China-specific or offer an alternative language presentation. This creates a natural-language locale constraint that may violate policy when users have not opted into Chinese-language operation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill operationalizes credentialed upload behavior as part of the normal workflow without a strong guardrail distinguishing local rendering from external publication steps. That makes accidental data exfiltration more likely, especially if the user only asked for formatting or draft preparation rather than remote upload.

Content

No source excerpt is available for this finding.