Back to skill

Security audit

skill-publisher

Security checks across malware telemetry and agentic risk

Overview

The skill is a plausible publisher, but it can delete local files, change global developer settings, patch installed tooling, and publish broadly without enough user control.

Install only if you are prepared to review every action first. Run a dry run, use a disposable staging copy of the skill folder, restrict markets explicitly, avoid passing tokens visibly on the command line, and do not allow it to change global git config or patch installed CLI files.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (16)

Lp3

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding
The skill reads local files such as SKILL.md and traverses the target directory, but it declares no permissions or equivalent user-facing notice for file access. In a publishing skill, local file reads are expected, yet the lack of explicit declaration reduces transparency and makes it easier to overreach into unintended paths.

Tp4

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding
The description emphasizes publishing and validation, but the implementation also performs destructive cleanup and other side effects that go beyond passive publishing assistance. Users invoking a broad 'publish skill' action may not reasonably expect local deletion behavior, creating a risk of data loss and unsafe consent.

Context-Inappropriate Capability

Medium
Confidence
96% confidence
Finding
The skill runs git config --global to overwrite the user's global Git identity, affecting unrelated repositories and future commits outside this task. A marketplace publisher does not need to mutate global developer configuration, so this is an unjustified persistent side effect.

Context-Inappropriate Capability

High
Confidence
98% confidence
Finding
The skill searches for an installed ClawHub CLI script and patches publish.js in place with sed before publishing. Modifying installed tool code on disk is a strong integrity violation that can break tooling, mask defects, or create a supply-chain style persistence risk beyond the current task.

Intent-Code Divergence

Medium
Confidence
83% confidence
Finding
The LobeHub step reports success for publication while the command shown is a local add/install style action or merely prints manual PR guidance. This can mislead users into believing a marketplace publish occurred when it did not, undermining auditability and causing accidental exposure or operational mistakes.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The README states that the publisher will remove files and that the cleanup step is conservative, but it does not give an explicit, prominent warning that user data may be deleted or excluded before publish. In a skill that operates on arbitrary local folders, destructive cleanup without clear consent, preview, backup guidance, or strong dry-run/default safeguards can cause accidental data loss or unintended repository modification.

Natural-Language Policy Violations

Medium
Confidence
83% confidence
Finding
The skill documentation says it will ensure README.zh-cn.md exists and presents Chinese README generation as part of the default publish pipeline rather than an opt-in choice. This can cause unintended disclosure or propagation of project content into another language and may create compliance, privacy, or accuracy issues for users who did not intend multilingual publication.

Missing User Warnings

Medium
Confidence
87% confidence
Finding
The README states the skill will create, refresh, and remove files, including stripping dotfiles, credentials, and other content, without a prominent warning about destructive behavior or overwrite risk. In a publishing skill that operates on local directories, unclear destructive semantics can lead to accidental deletion of important files or credential material before the user understands the impact.

Vague Triggers

Medium
Confidence
89% confidence
Finding
The trigger phrases are broad, including generic prompts like '发布 skill' or 'ship to marketplace,' which can activate the skill in contexts where the user did not intend multi-market publishing. Because this skill performs destructive cleanup, git mutation, and network publication, overbroad invocation increases the chance of harmful accidental execution.

Missing User Warnings

High
Confidence
97% confidence
Finding
The cleanup step recursively finds and deletes directories and files matching broad patterns like .env, *.pem, credentials.*, and node_modules under the supplied skill_dir. Without a prominent warning and explicit confirmation, a user can lose important local data or credentials simply by invoking a publishing workflow.

Missing User Warnings

Medium
Confidence
93% confidence
Finding
The skill automatically creates and pushes git tags to origin, but this remote mutation is not surfaced as a prominent warning in the high-level description. Users may unintentionally publish version markers to shared repositories, triggering CI/CD, releases, or team workflows without informed consent.

Missing User Warnings

Medium
Confidence
90% confidence
Finding
The skill transmits repository identifiers and token-authenticated requests to multiple external services, but it lacks a clear privacy and data-sharing warning. In a multi-market publisher, this broad external distribution materially increases exposure of metadata and possibly package contents if the user does not understand where data is going.

Missing User Warnings

Medium
Confidence
98% confidence
Finding
The document instructs users to run a remote installer via `curl ... | bash`, which executes network-fetched code immediately without an integrity check, signature verification, or review step. In a skill-publishing context this is especially risky because users are primed to paste commands verbatim, so a compromised server, CDN, DNS path, or install script could lead to arbitrary code execution on the user's machine.

Missing User Warnings

Medium
Confidence
96% confidence
Finding
The guide tells users to place a long-lived API token directly on the command line, which can leak through shell history, process listings, terminal logging, screenshots, or shared session tooling. Because this token authorizes publishing actions, exposure could let an attacker impersonate the user, publish malicious updates, or tamper with marketplace content.

Tool Parameter Abuse

High
Category
Tool Misuse
Content
RM_LIST=".git .DS_Store .idea .vscode node_modules .env .env.* *.key *.pem credentials.* secrets.* *.log *.tmp *.bak"

echo "=== 将删除 ==="
for item in $RM_LIST; do find {skill_dir} -name "$item" 2>/dev/null; done

[ "$DRY_RUN" = "true" ] && echo "[DRY RUN]" && exit 0
Confidence
97% confidence
Finding
The cleanup logic uses an unquoted user-controlled {skill_dir} inside find commands combined with wildcard name matching, enabling path abuse or unintended traversal if the parameter is malformed or points somewhere broader than expected. In context, this feeds directly into later deletion operations, so misuse can escalate to destructive removal of arbitrary files under the resolved path.

Tool Parameter Abuse

High
Category
Tool Misuse
Content
[ "$DRY_RUN" = "true" ] && echo "[DRY RUN]" && exit 0

for item in $RM_LIST; do
    find {skill_dir} -name "$item" -type d -exec rm -rf {} + 2>/dev/null
    find {skill_dir} -name "$item" -type f -delete 2>/dev/null
done
echo "✅ 清理完成"
Confidence
99% confidence
Finding
This line executes rm -rf and file deletion on paths returned by find under a user-supplied directory, creating a direct destructive primitive. If skill_dir is incorrect, attacker-influenced, or resolves outside the intended project, the skill can recursively delete unrelated directories, sensitive files, or development assets.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.