Back to skill

Security audit

Fire & Smoke Detection Skill | 火情烟雾检测技能

Security checks for vulnerabilities and agentic risk

Overview

This cloud fire-analysis skill needs review because it can mix users’ report accounts and process footage despite an export-only option.

Wait for account isolation and export-only behavior to be fixed before using this with sensitive footage or a shared agent. Confirm the provider’s report access and retention policies, how stored tokens can be removed, and that API secrets will not be reused as account identifiers. Treat its results as supplementary fire-warning information.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/fire_detection_analysis.py:55
Finding

Upstream sender identity is discarded in favor of a shared default report account

Content
View full analysis

Impact

A second sender can retrieve reports previously associated with the shared fallback account, including reports created during another sender's session. Analysis requests can also be attributed to the wrong account.

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
Findings (24)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied chunk implements account-data persistence rather than image or video analysis, flame/smoke inference, or fire alerts. Database storage alone could be supporting infrastructure, but the concrete user-profile, token, and default-account management capabilities are not explained by the declared detection purpose. No network transmission or token exfiltration is shown. This assessment is limited to the supplied chunk, not the complete skill.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Ordinary JSON, file, polling, and HTTP helpers could support a detection skill and are not independently mismatches. However, this chunk also implements substantial identity provisioning, credential persistence, and paid-platform account behavior that the description does not disclose. The request wrapper accepts absolute URLs and attaches configured credentials without checking the destination origin, creating an additional undeclared credential-disclosure capability. No flame/smoke analysis or detection-specific API call appears in the supplied chunk. Because this is a shared utility module, that absence alone does not establish that the complete skill lacks detection functionality; the mismatch rests on the additional account and credential behaviors shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The supplied code does not substantiate the declared detection purpose. The only explicit operational method concerns agent chat and is nonfunctional as written; it does not execute the commented-out external command. Imported base classes and API services could contain additional functionality, but their implementations are not supplied, so this assessment is limited to the visible chunk. No malicious behavior or active external command execution is demonstrated.

Content

No source excerpt is available for this finding.

YARA rule 'agent_skill_mcp_tool_poisoning_metadata': MCP/tool metadata poisoning indicators in tool schemas or skill manifests [agent_skills]

High
Category
YARA Match
Confidence
80% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: "fire-detection-analysis"
description: "Real-time detection of flames and smoke in video and image scenes. Suitable for fire early warning in industrial parks, forests, warehouses, and other locations. | 火情烟雾检测技能,实时检测视频/图片场景中的火焰、烟雾,适用于工业园区、森林、仓库等场所火情预警"
license: "MIT-0"
---

This capability supports real-time detection of flames and smoke in video streams or images, making it suitable for high
fire-risk locations such as industrial parks, forests, and warehouses. Based on color features, dynamic texture
analysis, and smoke dif

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, file_write, network, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The analysis workflow lacks a privacy warning about sending surveillance media to an external API and maintaining cloud-accessible reports.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

Although the documentation mentions APIs and cloud reports, it does not clearly disclose that local surveillance media is uploaded externally or explain retention and report-access controls. Industrial and warehouse footage may expose people, facility layouts, or operational details, making undisclosed processing a meaningful privacy risk.

Content

No source excerpt is available for this finding.

The --export-env-only help promises not to execute analysis, but the flag is never checked.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The export-only flag is parsed but never checked, so supplying it alongside an input still initializes identity and runs analysis. A caller relying on the explicit no-analysis promise can unintentionally upload sensitive surveillance media or trigger remote processing.

Content

No source excerpt is available for this finding.

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · skills/smyx_analysis/scripts/skill.py (reported line 28)May include surrounding context.

python
result_json = JsonUtil.parse(result_json_pure_text, result_json_pure_text)

        result_json_common_ai_response = result_json.get("commonAiResponse") if isinstance(result_json,
                                                                                           dict) else result_json
        if result_json_common_ai_response:
            result_json = result_json_common_ai_response

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · skills/smyx_analysis/scripts/skill.py (reported line 33)May include surrounding context.

python
result_json = result_json_common_ai_response

        result_json_health_ai_response = result_json.get("healthAiResponse") if isinstance(result_json,
                                                                                           dict) else result_json
        if result_json_health_ai_response:
            result_json = result_json_health_ai_response

Local video contents are submitted to an API without disclosure of the upload.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The implementation reads the complete local media file and passes it to the API analysis call without an upload-consent or approved-policy check. Surveillance footage can contain sensitive personal and operational information, and a request to analyze a local file does not necessarily authorize undisclosed external transfer.

Content

No source excerpt is available for this finding.

The --export-env-only option promises not to execute analysis, but analysis still runs.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The export-only option promises not to analyze, but its value is never consulted before identity resolution and the normal analysis path. With an input supplied, this can cause an unexpected external media upload despite the caller selecting a non-analysis mode.

Content

No source excerpt is available for this finding.

The skill includes persistent user-profile and credential management unrelated to flame/smoke detection.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

L448–L465 define a system-user table containing personal information and authentication tokens, and L470–L473 expose it through the general CRUD implementation. DAO initialization also creates and migrates this table (L180–L214), so this is an implemented persistence capability, not merely unused field documentation. Managing user identities and credentials is not justified by the manifest's stated video/image fire-detection and early-warning purpose.

Content

No source excerpt is available for this finding.

The ai_chat docstring claims to invoke an OpenClaw agent, but the implementation never invokes it or returns analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The docstring at L26 states that this method calls the OpenClaw agent through subprocess, and L33-L34 promise an analysis result or session ID. However, command construction and subprocess execution are commented out; L68 initializes an empty dictionary, and accessing result.stderr at L70 raises an AttributeError that is caught at L81, causing an implicit None return instead of the documented operation.

Content

No source excerpt is available for this finding.

FileUtil.open silently truncates existing files without disclosing its write-only behavior.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

Opening a path with mode 'w' immediately erases any existing contents. This generic open helper has no docstring, comment, confirmation, or action log explaining that destructive behavior; exception tracing only runs after an error.

Content

No source excerpt is available for this finding.

Detection requests can silently register a remote user and persist account credentials, although the manifest describes only flame and smoke detection.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

An ordinary request can automatically call a registration endpoint with silent=1 and register=1, then persist returned account tokens without a separate enrollment decision. A detection-only description does not disclose this external account creation and credential lifecycle, and the selected username can also originate from API_SECRET_KEY.

Content

No source excerpt is available for this finding.

HTTP requests can silently trigger remote user registration without disclosure.

Medium
Category
Output Handling
Confidence
95% confidence
Finding

The generic HTTP wrapper hides a state-changing registration operation behind routine requests when credentials are missing. This can create an external account and disclose an identifier without a distinct enrollment choice, an unexpected side effect for a flame/smoke detection skill.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The POST sends the selected identity as both openId and mobile to a configured remote login endpoint. Because the caller preferentially selects API_SECRET_KEY as that identity, the request can transmit a credential in registration fields, potentially exposing it to account records or logging systems beyond its intended authentication use; this does not establish an attacker-controlled destination.

Content

Scanner excerpt · skills/smyx_common/scripts/util.py (reported line 561)May include surrounding context.

python
"source": ConstantEnum.DEFAULT__SKILL_HUB_NAME
            }
            try:
                _response = requests.post(_url, json=_data)
                if _response.status_code == 200:
                    _response_json = _response.json()
                    if _response_json and _response_json.get("success"):

The user-facing analysis heading is fixed in Chinese without a language choice or opt-in.

Low
Category
Output Handling
Confidence
90% confidence
Finding

The output-heading method unconditionally returns a Chinese title, regardless of the user's preferred language. The file provides no language selection or documented justification for this locale constraint.

Content

No source excerpt is available for this finding.

Configuration loading silently creates directories and YAML files without disclosing the filesystem changes.

Low
Category
Output Handling
Confidence
94% confidence
Finding

When a configuration file is missing, YamlUtil.load creates its parent directories and writes a new file without a log, confirmation, docstring, or explanatory comment. BaseEnum.init_subclass invokes this loader at L053 and L058, so these filesystem changes can occur during module import rather than an explicit save operation.

Content

No source excerpt is available for this finding.

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · skills/smyx_common/scripts/dao.py (reported line 243)May include surrounding context.

python
username = getattr(model, "username", None)
            if username:
                column_names = self.__model__.__table__.columns.keys()
                update_data = {key: getattr(model, key) for key in column_names if key != "username"}
                return self.update_by_username(username, **update_data)

            return None

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · skills/smyx_common/scripts/dao.py (reported line 326)May include surrounding context.

python
if filters:
                for key, value in filters.items():
                    query = query.filter(getattr(self.__model__, key) == value)

            if offset:
                query = query.offset(offset)

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · skills/smyx_common/scripts/dao.py (reported line 441)May include surrounding context.

python
if filters:
                for key, value in filters.items():
                    query = query.filter(getattr(self.__model__, key) == value)

            return query.scalar()
        finally:

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · skills/smyx_common/scripts/dao.py (reported line 353)May include surrounding context.

python
column_names = self.__model__.__table__.columns.keys()

            for key in column_names:
                value = getattr(model, key)
                setattr(instance, key, value)

            session.commit()

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · skills/smyx_common/scripts/util.py (reported line 30)May include surrounding context.

python
elif isinstance(cur, dict):
                        cur = cur.get(part, default)
                    else:
                        cur = getattr(cur, part)
                return cur
            except Exception:
                return default