Back to skill

Security audit

EvoMap WorkBench v1.0.11 Mini

Security checks for vulnerabilities and agentic risk

Overview

This skill is a Review risk because it is advertised as a local AI workbench but includes under-disclosed code that can reuse local Feishu credentials and send data externally.

Review this skill carefully before installing. It may be useful as an EvoMap decision workbench, but only install it in an environment where you are comfortable with it reading OpenClaw Feishu credential/config files and where outbound notification and validation traffic is explicitly intended. Avoid using it with shared production Feishu credentials unless the credential source, recipients, and transmitted fields have been reviewed. Back up OpenClaw configuration before running the uninstaller.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (4)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
lib/notification_system.py:75
Finding

Automatic Cross-Skill Credential Discovery Enables Confused-Deputy Message Delivery

Content
View full analysis
Dict: if config_file: with open(config_file, 'r', encoding='utf-8') as f: return json.load(f) default_paths = [ '/home/admin/.openclaw/workspace/.config/feishu-notification.json', '/home/admin/.openclaw/credentials/feishu-config.json', '/home/admin/.openclaw/credentials/feishu-pairing.json' ] config = {} for path in default_paths: if Path(path).exists(): with open(path, 'r', encoding='utf-8') as f: config.update(json.load(f)) if not config.get('app') and not config.get('webhook'): openclaw_feishu_paths = [ '/home/admin/.openclaw/workspace/.config/python-learning-state.json', '/home/admin/.openclaw/credentials/feishu-default-allowFrom.json', '/home/admin/.openclaw/credentials/feishu-pairing.json' ] for path in openclaw_feishu_paths: if Path(path).exists(): with open(path, 'r', encoding='utf-8') as f: openclaw_config = json.load(f) if 'appSecret' in openclaw_config: if 'app' not in config: config['app'] = {} config[ ...[truncated 4187 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
lib/notification_system_compliant.py:182
Finding

The “Compliant” Notification Variant Retains Automatic Credential Access and Hard-Coded Delivery

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
lib/asset_validator.py:94
Finding

Remote Asset Validation Uploads Complete Assets and Bearer Credentials Without Data-Minimization Controls

Content
View full analysis
Tuple[bool, str, Optional[str]]: url = "https://evomap.ai/a2a/validate" headers = { "Authorization": f"Bearer {auth_token}", "Content-Type": "application/json" } payload = { "payload": { "assets": [asset] } } try: resp = requests.post( url, json=payload, headers=headers, timeout=10 ) data = resp.json() if resp.status_code == 200 and data.get('status') == 'ok': computed_assets = data.get('computed_assets', []) if computed_assets: computed_id = computed_assets[0].get('computed_asset_id') return True, "Hub 验证通过", self.ensure_sha256_prefix(computed_id) else: return False, "Hub 未返回 computed_assets", None else: error_msg = data.get('msg', data.get('error', '未知错误')) return False, f"Hub 验证失败:{error_msg}", None except requests.exceptions.Timeout: return False, "Hub 验证超时", None except Exception as e: return False, f"Hub 验证异常:{str(e)}", None ``` ### Technical Analysis The local validator can calculate the asset hash itself, but `validate_with_hub()` uploads the complete asset object to `https://evomap.ai` along with a bearer token. There is no field allowlist, redaction step, payload-size limit, sensitivity check, user confirmation, or policy hook. Assets may contain more than the fields needed to calculate or validate a hash. Because the complete dictionary is serialized, incidental secrets, private prompts, customer data, internal URLs, or ...[truncated 1571 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
uninstall.py:109
Finding

Uninstaller Deletes Broadly Name-Matched Configuration Files and Credential Fields

Content
View full analysis
bool: credentials_path = self.base_path / "credentials" / "feishu-pairing.json" if not credentials_path.exists(): print("✅ feishu-pairing.json 不存在,无需恢复") return True config = self.load_json_file(credentials_path) evomap_keys = [ "evomap_appId", "evomap_appSecret", "cli_a929676f8bf81cc7" ] keys_to_check = [] for key in config.keys(): for evomap_key in evomap_keys: if evomap_key.lower() in key.lower(): keys_to_check.append(key) if keys_to_check and remove_app_config: for key in keys_to_check: del config[key] if self.save_json_file(credentials_path, config): return True return False ``` ```python def remove_skills_config(self) -> bool: config_dir = self.base_path / "config" / "skills" if not config_dir.exists(): return True evomap_configs = [] for config_file in config_dir.iterdir(): if config_file.is_file() and ( "evomap" in config_file.name.lower() or "workbench" in config_file.name.lower() ): evomap_configs.append(config_file) if evomap_configs: for config_file in evomap_configs: try: config_file.unlink() except Exception as e: print(f"⚠️ 无法删除 {config_file.name}: {e}") return True ``` ### Technical Analysis The uninstaller edits a shared credential file by deleting every top-level key whose name merely contains one of three substrings. It also deletes every file in the shared Skill configuration directory whose filename contains `evomap` or `workbench`. These checks ...[truncated 1690 chars]
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (83)

YARA rule 'agent_skill_credential_exfiltration_webhook': AI agent skill credential harvesting followed by webhook or external exfiltration [agent_skills]

Critical
Category
YARA Match
Confidence
99% confidence
Finding

The YARA pattern is strongly supported by the code flow: it reads host credential files and environment secrets, then uses those credentials to authenticate and send data across several external messaging channels. In an agent skill, that combination is highly dangerous because it can convert ambient host secrets into covert outbound communication, matching credential harvesting followed by exfiltration behavior.

Content

Scanner excerpt · lib/notification_system.py (reported line 132)May include surrounding context.

python
if allow_from:
                                config['targetUser'] = allow_from[0]
                                print(f"[飞书] ✅ 已从 OpenClaw 配置加载允许的用户")
                        
                        break
        
        # 如果还是没有配置,尝试从环境变量加载
        if not config.get('app'):
            import os
            app_id = os.environ.get('FEISHU_APP_ID')
            app_secret = os.environ.get('FEISHU_APP_SECRET')
            
            if app_id and app_secret:
                config['app'] = {
                    'appId': app_id,
                    'appSecret': app_secret
                }
                print(f"[飞书] ✅ 已从环境变量加载配置")
        
        return config
    
    def _get_access_token(self) -> str:
        """获取访问令牌"""
        if self.access_token and time.time() < self.token_expire:
            return self.access_token
        
        # 检查 app_secret 是�

YARA rule 'agent_skill_credential_exfiltration_webhook': AI agent skill credential harvesting followed by webhook or external exfiltration [agent_skills]

Critical
Category
YARA Match
Confidence
99% confidence
Finding

The combination of automatic credential harvesting from local files/environment variables and multiple outbound webhook/API send paths strongly matches a credential-exfiltration pattern. Even if framed as convenience or compatibility, this design enables the skill to appropriate existing secrets and use them to communicate externally without a narrowly justified relationship to the advertised workbench purpose.

Content

Scanner excerpt · lib/notification_system_compliant.py (reported line 98)May include surrounding context.

python
添加 search_only 标志(如果请求体中包含 EvoMap API 相关字段)
    if 'payload' in payload and isinstance(payload['payload'], dict):
        # 默认添加 search_only: true,避免不必要的扣费
        if 'search_only' not in payload['payload']:
            payload['payload']['search_only'] = True
    
    for attempt in range(max_retries):
        try:
            resp = requests.post(url, json=payload, headers=headers, timeout=10)
            data = resp.json()
            
            # 处理 429 错误(限流)
            if resp.status_code == 429:
                global _last_429_time
                _last_429_time = datetime.now()
                print(f"[通知系统] ⚠️ 429 限流,等待 {RATE_LIMIT_DELAY} 秒后重试 (尝试 {attempt+1}/{max_retries})")
                
                if attempt < max_retries - 1:
                    time.sleep(RATE_LIMIT_DELAY)
                    continue
                else:
                    print(

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Undeclared uninstallation behavior that edits local registry/configuration files, deletes credential-related entries, removes skill configs, and recursively deletes directories is highly sensitive. In context, this is more dangerous because the skill is framed as an AI productivity component, not a system-maintenance utility, so users may not anticipate destructive file operations.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · install.py (reported line 277)May include surrounding context.

python
parser = argparse.ArgumentParser(description="EvoMap WorkBench v1.0.11 - 安装工具")
    parser.add_argument("--source", type=str, default=None,
                        help="源路径(默认为发布包位置)")
    parser.add_argument("--skip-check", action="store_true",
                        help="跳过环境检查")
    
    args = parser.parse_args()

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The code automatically scans OpenClaw credential/config paths and environment variables to harvest Feishu secrets and target identifiers, then normalizes them for later use. In an agent skill context, opportunistic secret discovery beyond explicit user input is dangerous because it can silently appropriate credentials from the host environment and enable unauthorized messaging or downstream exfiltration.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

This section acquires access tokens and sends messages to Feishu using externally supplied content and target user IDs. In the context of a skill not primarily described as a messaging tool, this unjustified network capability enables silent transmission of task data, results, or secrets to third-party endpoints.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The code automatically searches fixed OpenClaw credential/config paths and environment variables, then loads app secrets and user identifiers without an explicit user action. This is dangerous because it silently expands the trust boundary to unrelated local secrets and can repurpose existing credentials to contact external services.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill advertises no explicit tool scope or permissions, yet the associated behavior indicates access to environment variables, local files, file writes, and network. That combination creates a significant transparency and consent problem: users and the platform cannot accurately assess what resources the skill will touch, and hidden capability expansion can enable credential access, persistence, or data exfiltration.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest description and nearly all user-facing documentation are written in Chinese, with no indication that users may choose another language or that the skill is intentionally limited to a Chinese-speaking or region-specific audience. This creates a natural-language policy concern because the skill effectively imposes a language preference without user opt-in or justification.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The module docstring says the installer will '自动处理 3 个 JSON 文件的修改', and the json_files_to_process mapping labels skill_metadata.json as 'update' and feishu-pairing.json as 'optional_read_write'. In practice, process_skill_metadata only reads and prints metadata, read_feishu_config only reads credentials, and only update_skills_index writes a JSON file, so the inline documentation actively overstates modification behavior.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The installer reads a credential file containing Feishu app identifiers and secrets even though those secrets are not required for the core file-copy and registration steps. Accessing unrelated credentials expands trust scope, creates unnecessary exposure of sensitive data in memory/logging paths, and violates least-privilege expectations for an installer.

Content

No source excerpt is available for this finding.