Back to skill

Security audit

Automatic Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed skill-generation and publishing pipeline, but it needs review because it can run generated code and publish public artifacts using the user's credentials without strong containment.

Install only if you are comfortable with an agent generating and executing code on your machine and publishing it under your GitHub and ClawHub accounts. Use a disposable workspace, keep publishing tokens out of the environment until the upload step, prefer dry-run and PR workflow, review generated package.json and scripts before npm install or execution, and inspect the exact files to be published before any public push or clawhub publish.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (4)

T08 · Insecure Dependencies

Error
Location
scripts/self-run.js:43
Finding

Unsafe Installation of Dependencies from Generated Packages

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
scripts/pipeline.js:68
Finding

Generated Scripts Execute with Publishing Credentials Before the Safety Gate

Content
View full analysis
= 18 is installed ``` The self-run stage then executes every generated script: ```text STEP C — Run each script For each script, run it with no arguments first: node ${skillDir}/scripts/.js Expected results: - Scripts that require args: should print a usage message and exit with code 1 (PASS) - Scripts that work without args: should print a no ...[truncated 2721 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/upload.js:112
Finding

Command Injection and Unsafe Deletion Through Unvalidated Pipeline Values

Content
View full analysis
"` : `If NOT_FOUND: # Read the tagline from SKILL.md description field (first line after "description: |") gh repo create ${githubOwner}/${slug} --public --description "" If EXISTS: skip creation, proceed to STEP 3.`} STEP 3 — Push skill files to standalone repo${dryRun ? ' (SKIPPED — dry-run)' : ''} ${dryRun ? 'DRY-RUN: would initialize standalone repo and push skill files' : `Create a clean local git repo from the skill directory and push: cd /tmp rm -rf ${slug}-standalone cp -r ${skillDir} ${slug}-standalone cd ${slug}-standalone git init git add . git commit -m "feat: ${slug}@${version} — auto-generated by automatic-skill" git remote add origin https://github.com/${githubOwner}/${slug}.git git push -u origin main Verify standalone push: gh api repos/${githubOwner}/${slug}/commits/main --jq '.sha + " | " + .commit.message' → Should show the new commit`} ``` Relevant values are derived from command-line input, pipeline state, or environment data without strict validation: ```js if (!skillDir) { skillDir = args.filter(a => a !== '--lang' && a !== lang && !a.startsWith('--'))[0] || ''; slug = path.basename(skillDir); } const githubRepo = process.env.GITHUB_REPO || ''; const branchName = `auto-skill/${slug}`; const githubOwner = githubRepo.includes('/') ? githubRepo.split('/')[0] : ''; ``` ### Technical Analysis The script is a prompt generator, but its output explicitly directs the Agent to run shell commands. Values such as `slug`, `ski ...[truncated 2009 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/scan-check.js:86
Finding

Incomplete Secret Scanning Allows Unreviewed Files to Be Publicly Published

Content
View full analysis
{ if (trigger.regex.test(line)) { ``` The upload stage then copies and stages the complete directory: ```text cd /tmp rm -rf ${slug}-standalone cp -r ${skillDir} ${slug}-standalone cd ${slug}-standalone git init git add . git commit -m "feat: ${slug}@${version} — auto-generated by automatic-skill" git remote add origin https://github.com/${githubOwner}/${slug}.git git push -u origin main ``` ### Technical Analysis The scanner does not recursively traverse the complete Skill tree. It examines only: - Files directly under the Skill root. - Files directly under `scripts/`. - Files ending in `.js` or `.md`. It therefore ignores common sensitive or executable content, including: - `.env` and other dotfiles. - JSON, YAML, TOML, XML, shell scripts, and configuration files. - Private keys and credential stores. - Nested directories other than the first level of `scripts`. - Archives, binaries, generated artifacts, and symlink targets. In contrast, `cp -r` and `git add .` include the entire directory. The publication operation theref ...[truncated 1611 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (72)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

声明描述的是一个端到端、零人工介入、可定时自动运行并发布产物的完整自动化工厂;但代码块只是一个生成“创建阶段提示词”的 Node.js 脚本。它依赖先前阶段产物(design/seo),从环境变量和 current-pipeline.json 读取数据,然后 console.log 输出指令文本,要求外部代理去创建文件、更新流水线并进入 review。代码没有显示任何定时器、网络发布、GitHub/clawHub API 交互、测试执行、验证逻辑或多阶段编排,因此其实际能力明显窄于声明的主要用途,属于实质性描述不符。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The declared description presents a comprehensive autonomous skill factory with end-to-end research, creation, testing, upload, verification, and publishing behavior. The supplied code chunk does not implement that primary behavior. Instead, it is a wrapper/entrypoint that checks lock files, prevents duplicate runs, creates a lock, and prints instructions indicating what should happen next. There is no actual child-process execution of pipeline.js, no stage orchestration logic, no GitHub or clawHub interaction, no logging to pipeline-log.json, and no notification sending. Some of the declared behavior could exist elsewhere, but based on this chunk alone, the behavior is materially narrower than the description, so this is a mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

There is a strong description-to-behavior mismatch. The declared description presents a complete autonomous production pipeline with scheduling, multi-stage orchestration, generation, testing, publishing, and upgrade flows. The actual code chunk is much narrower: it is a prompt generator for the design stage only. It reads command-line input or a selected idea from a local JSON file and prints a specification prompt in Chinese or English. Importantly, despite the prompt text saying to read/update current-pipeline.json and proceed to Stage 3, the script itself does not do those actions—it only outputs those instructions. This is not just an incomplete subset of a larger capability claim; the code’s primary behavior is materially different from the declared end-to-end autonomous factory behavior.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The declared description presents a complete autonomous system that can independently run all 10 stages, schedule daily runs, generate and publish production-ready skills, and support manual triggering or stage-specific debugging. The supplied code chunk, however, is a narrow script for the final review stage only. Its concrete behavior is limited to parsing arguments, optionally reading current-pipeline.json to derive a slug, and printing a detailed prompt/instruction block describing what an agent should do next. It does not itself carry out the promised end-to-end workflow, nor does it actually write pipeline-log.json, delete current-pipeline.json, upload artifacts, verify publishing, or perform scheduling. Because the primary purpose and implemented capabilities of this code chunk are materially narrower than the declared description, this is a mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The declared description claims a comprehensive automatic skill factory with a 10-stage production pipeline and multiple operational modes. The actual code chunk is much narrower: it parses on|off|status arguments and outputs textual guidance for using openclaw cron commands. It does not modify cron state itself, does not invoke the pipeline, and does not implement any of the major capabilities in the description except loosely referencing a daily pipeline script and the 02:00 schedule. This is a material description/behavior mismatch because the actual primary purpose of the code is a cron toggle helper, not the autonomous skill factory described.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The description claims a production-grade autonomous skill factory with 10 stages, scheduled daily runs, manual triggering, iteration of existing skills, and publication to GitHub and clawHub. The supplied code chunk is much narrower: it is a single script for Stage 1 that merely outputs instructions/prompts to stdout. It does not execute searches, inspect directories, write files, call other stages, schedule jobs, interact with GitHub/clawHub, or perform any end-to-end automation. This is a material description-versus-behavior mismatch, not just an incomplete implementation detail, because the declared primary purpose is broad autonomous orchestration while the actual code shown is only a research prompt template.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared description presents a broad end-to-end autonomous factory that can research, design, create, test, upload, verify, and publish skills on a daily schedule. The actual code chunk is much narrower: it is a command-line script for Stage 5 Review that reads an optional pipeline file to locate a skill directory and prints a review checklist/prompt in Chinese or English. It does not execute the review, does not update pipeline JSON, and does not perform any of the major factory functions claimed in the description such as scheduling, generation, self-running, self-checking, upload, verification, or publishing. This is a material description-behavior mismatch rather than a mere partial implementation detail.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The description claims a broad autonomous factory that can generate, test, and publish production-ready skills across a 10-stage pipeline. The actual code shown implements a narrow Stage 7.5 quality gate: scanning local files for potential credential leaks before upload and printing results. While this safety check could be a supporting component within such a pipeline, the supplied code chunk does not itself exhibit the primary declared behavior. There are no network operations, no scheduling, no content generation, no orchestration across stages, and no publishing logic. Therefore the declared description materially overstates and misrepresents what this code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description promises a production-grade automated skill factory with end-to-end generation, testing, and publication workflows, including scheduled runs and deployment targets. The supplied code does none of that. It only parses an optional --lang argument and prints documentation about credential-safe scripting practices in Chinese or English. The code explicitly states it is a prompt/documentation generator only, makes no outbound network requests, does not modify files, and does not re-publish. This is a clear material mismatch in primary purpose and capabilities.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The declared description covers a broad autonomous production pipeline for generating and publishing skills. The actual code chunk is much narrower: it parses arguments, optionally reads a skill directory from a pipeline JSON file, and prints a detailed English or Chinese self-check prompt. There is no implementation of the 10-stage workflow, no scheduler, no network or publishing behavior, no GitHub/clawHub interaction, and no actual file validation or JSON report generation beyond instructions in the printed text. This is a material description-behavior mismatch, not just an incomplete snippet of a larger system, because the supplied code's actual behavior is limited to emitting instructions for Stage 7.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The description presents a broad autonomous skill-generation and publishing system with daily scheduling, manual triggering, stage execution, upgrades, SEO optimization, and publication to GitHub/clawHub. The supplied code chunk, however, is only a status/history management script. Its primary function is to inspect local pipeline state files, print progress/history, and optionally clear current state and a lock file. That is a materially narrower and somewhat different behavior than the declared end-to-end autonomous factory. Additionally, the declared 10-stage pipeline includes SEO, but this script’s stage map contains only 9 stages and omits SEO. While a status script could be a supporting component of the larger described system, this specific chunk includes an undeclared administrative capability (--clear) and does not itself implement the main declared functionality.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

声明描述的是一个端到端、全自动、定时执行并可发布的 skill 工厂;但代码片段实际只是“上传阶段”的提示生成器。它虽然与声明中的 upload 子功能相关,但主能力明显缩小:脚本只根据本地文件和环境变量拼接出供后续 agent/人工执行的命令说明,并特别声明自己不发起网络请求。由于用户要求核对‘描述是否准确代表代码实际行为’,这里描述显著夸大了代码片段的范围与自动化程度,因此构成不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

描述强调的是一个端到端、零人工介入的自动 skill 工厂;而实际代码块只是流水线中“验证上传”阶段的说明生成器。它读取本地 JSON 状态,拼装 slug、版本、commit 信息,然后将应由外部代理或用户执行的 gh/clawhub 命令打印出来。代码注释还明确写着“PROMPT GENERATOR ONLY — this script makes NO outbound network requests”。因此,这段代码的实际能力远小于声明的整体产品能力,且其主要目的也与“全自动工厂”不一致。虽然如果把描述理解为整个 skill 仓库而非单个脚本,代码可能是其中一个子阶段组件,但就所给代码块而言,存在明显的描述—行为不匹配。

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
93% confidence
Finding

The skill requires access to high-value credentials such as GITHUB_TOKEN and CLAWHUB_TOKEN and describes write-capable use of those tokens for repo creation, pushes, and publishing. Credential access is especially sensitive here because the skill is framed as autonomous and can combine token use with file generation and public release actions, so misuse or prompt-trigger mistakes could cause unauthorized publication or repository modification.

Content

Scanner excerpt · SKILL.md (reported line 58)May include surrounding context.

md
env:
    - name: GITHUB_TOKEN
      required: true
      description: "GitHub personal access token with repo write permission (narrowly scoped to a single repo)."
    - name: GITHUB_REPO
      required: true
      description: "Target repository in owner/repo format, e.g. myorg/openclaw-skills"

Credential Access

High
Category
Privilege Escalation
Confidence
95% confidence
Finding

The metadata explicitly requires GitHub and ClawHub tokens for write operations, creating a real security boundary around credentialed publishing. Even if the scripts are only prompt generators, the skill is designed to orchestrate privileged actions through external CLIs, so accidental or malicious downstream use could modify repositories or publish content under the user's accounts.

Content

Scanner excerpt · SKILL.md (reported line 89)May include surrounding context.

md
env:
      - name: GITHUB_TOKEN
        required: true
        description: "GitHub personal access token — must have 'repo' scope. Scoped to your own repos only. Used by gh CLI to create repos and push commits."
      - name: GITHUB_REPO
        required: true
        description: "Target monorepo in owner/repo format, e.g. yourname/openclaw-skills. Must be a repo you own."

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 166)May include surrounding context.

md
- 查看流水线状态和历史日志(`status.js`)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 350)May include surrounding context.

md
- 查看流水线状态和历史日志(`status.js`)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 351)May include surrounding context.

md
- 查看流水线状态和历史日志(`status.js`)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 352)May include surrounding context.

md
- 查看流水线状态和历史日志(`status.js`)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 275)May include surrounding context.

md
| 1 | Research | `research.js` | Scan trends, identify skill gaps, output Top-3 ideas |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 337)May include surrounding context.

md
| 1 | Research | `research.js` | Scan trends, identify skill gaps, output Top-3 ideas |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 277)May include surrounding context.

md
| 3 | SEO | `seo.js` | Optimize display name, tagline, description, 30+ keywords |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 339)May include surrounding context.

md
| 3 | SEO | `seo.js` | Optimize display name, tagline, description, 30+ keywords |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 293)May include surrounding context.

md
Every skill generated or uploaded by automatic-skill **must** include all five of the following sections in its `SKILL.md`, in this order:

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 332)May include surrounding context.

md
node scripts/pipeline.js # auto-select topic, run all stages