Back to skill

Security audit

Time-Lapse Album Custom Summarization Skill | 时光相册自定义浓缩分析技能

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with high confidence because scanners report behavior beyond its video-highlights purpose, including persistent account and credential handling, and flaws in its analysis and upload controls.

Consider avoiding installation unless you are comfortable with the reported remote video submission, account and token handling, and local persistence. Scanning was partial because a referenced local path was missing, so the evidence is incomplete; the scanners did not establish malicious intent.

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
  • 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 (27)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied chunk implements user-account persistence rather than video analysis, target-segment extraction, or highlight compilation. Database support can be a legitimate implementation detail, but this chunk specifically manages personal profiles and authentication-token fields without any demonstrated connection to the declared media-processing function. No network transmission or credential harvesting is shown; the mismatch is the unrelated user-management capability. This assessment is limited to the supplied code chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

analyze_timelapse() selects input_path or url or text and passes only that value to skill.get_output_analysis(). Since normal execution requires --input or --url, --text can never reach the downstream analysis call. Thus, the central promise of generating highlights based on specified keywords or targets is not represented by this code. The imported skill implementation is not supplied, so whether it performs generic video extraction or compilation cannot be verified. Identity resolution, result listing, and optional output-file writing appear to be supporting functionality rather than unrelated capabilities.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Ordinary formatting, polling, filesystem, and API-client helpers can support a video-highlights skill and are not inherently mismatches. However, this chunk also implements undeclared account provisioning and persistent identity/credential handling. Its request wrapper accepts absolute URLs and adds configured credentials without checking the destination, extending access beyond a narrowly scoped highlight-generation operation. These behaviors occur when the relevant helpers are invoked, not necessarily on import. No video analysis or compilation appears in this utility chunk, although that absence alone would not establish a mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

The visible implementation does not substantiate the declared purpose. It contains scaffolding and a nonfunctional chat helper, not a video-processing workflow. The commented-out command does not execute, so there is no evidence of actual subprocess invocation or external agent access. Imported base classes and services are not supplied; their behavior cannot be assessed, so this finding is limited to the provided chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
87% confidence
Finding

Video validation, file reading, submission, and polling could reasonably support the declared purpose and are not independently mismatches. However, the exposed outputs are structured reports and report-image links, and the code separately exposes report-history retrieval. Those capabilities are not represented in the description. The supplied chunk contains no visible segment extraction or highlight compilation; an unseen service might perform those operations, so their absence here alone is not conclusive.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
86% confidence
Finding

The code explicitly supports retrieving prior analysis records using an open ID, extending beyond generating highlights from a supplied video. Its analysis path only forwards a path or URL to skill.get_output_analysis; it accepts no keyword or target argument and contains no visible segment selection or compilation logic. Because the delegated implementation is not supplied, highlight generation cannot be ruled out, but it is not established by this chunk. Printing results and optionally saving them are ordinary supporting details, not independent mismatches.

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: "custom-timelapse-analysis"
description: "Generates condensed album highlights based on specified keywords or targets. It extracts specific target segments from long videos and compiles them into an exciting summary. | 时光相册自定义浓缩分析技能,基于指定关键词/目标生成时光相册浓缩集锦,从长视频中提取特定目标片段,整理生成精彩汇总"
version: "1.0.25"
license: "MIT-0"
---

# ⏱️ Time-Lapse Album Custom Summarization Skill | 时光相册自定义浓缩分析技能
> **智能分析中枢** · 图片/视频智能分析 · 结构化报告 · 历史报告云端查询

-

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 local-video analysis workflow lacks a clear disclosure that personal footage is sent to a remote analysis service.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The local-file workflow does not clearly tell users that the full video will be uploaded to an external analysis service. General references to APIs and cloud history are not an explicit upload disclosure, which matters particularly for family footage and videos of children.

Content

No source excerpt is available for this finding.

The custom target or keywords are discarded before analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The manifest promises highlights based on specified keywords or targets, but analyze_timelapse selects only one value from input_path, url, and text and passes it to the analysis backend. Because the CLI requires an input path or URL, the supplied target text is never forwarded, so the backend cannot perform the requested keyword-directed extraction through this interface.

Content

No source excerpt is available for this finding.

The documented environment-export-only mode does not prevent analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The CLI accepts --export-env-only and promises not to perform analysis, but never checks the parsed flag. With a valid input, execution still invokes analysis, potentially uploading private footage despite the caller explicitly selecting a non-analysis mode.

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 the analysis API without an upload disclosure.

Medium
Category
Output Handling
Confidence
93% confidence
Finding

The method reads the complete local video into memory and passes its contents to the analysis API without an upload-approval check. In a personal-video skill, silently crossing the local-to-remote boundary can expose sensitive household footage when users expect local processing.

Content

No source excerpt is available for this finding.

The documented --export-env-only mode still executes video analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The parser advertises an export-only mode that does not execute analysis, but main never branches on the flag. Supplying it alongside a valid video therefore still reaches the analysis client, potentially transmitting private content contrary to the caller's explicit choice.

Content

No source excerpt is available for this finding.

The skill includes persistent user-profile and credential management beyond its stated video-highlight generation purpose.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

The User model stores identifying information, demographic data, and token/open_token credentials (L448–L465), while UserDao exposes inherited read, update, and deletion operations (L470–L473). DAO initialization also creates and migrates the sys_user table (L180–L215). This is a persistent account-management capability, not video-segment analysis or highlight compilation, and the supplied manifest does not justify it.

Content

No source excerpt is available for this finding.

The documented agent invocation is disabled; ai_chat never performs analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The docstring at L25–L35 claims this method invokes openclaw agent through a subprocess and returns an analysis result or session ID. Instead, the subprocess invocation is commented out, and accessing stderr on the empty dictionary at L68–L70 always raises an exception that is caught, causing the method to return None without invoking an agent.

Content

No source excerpt is available for this finding.

FileUtil.open silently truncates existing files without disclosing its destructive write mode.

Medium
Category
Output Handling
Confidence
95% confidence
Finding

The generically named open helper always uses mode 'w', which immediately erases an existing file's contents. No docstring, comment, confirmation, or operation log discloses this behavior; exception tracing only runs after a failure.

Content

No source excerpt is available for this finding.

The request helper silently provisions remote user accounts and persists authentication credentials, beyond the stated video-highlights purpose.

Medium
Category
Excessive Agency
Confidence
92% confidence
Finding

A general HTTP request can implicitly trigger remote registration with silent=1 and register=1, then cache returned authentication credentials without a separate enrollment decision. More concretely, API_SECRET_KEY takes precedence as the username and can consequently be transmitted as both mobile and openId, unnecessarily exposing a secret through identity fields.

Content

No source excerpt is available for this finding.

HTTP requests can silently initiate remote login or registration without disclosing the identity transmission.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

When tokens are absent, ordinary request handling can send an identity to a remote login endpoint and request account creation automatically. Remote authentication may be legitimate for video processing, but coupling undisclosed enrollment to generic requests removes a meaningful consent boundary and can transmit an API secret when it is selected as the identity.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

The external POST is not inherently unsafe, but its payload includes username in both mobile and openId, while the caller preferentially derives username from API_SECRET_KEY. In that configuration, a secret is unnecessarily disclosed to the configured login service as profile data, potentially broadening its storage and logging exposure.

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"):

Loading a missing configuration silently creates directories and writes a file.

Low
Category
Output Handling
Confidence
93% confidence
Finding

YamlUtil.load creates parent directories and writes a YAML file when the requested path does not exist, without a comment, docstring, prompt, or log disclosing this side effect. BaseEnum.init_subclass invokes this loader during class creation (L044–L058), so merely importing the module can modify the filesystem rather than only read configuration.

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 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/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)