Back to skill

Security audit

KeplerJAI Bulletin Publish

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly coherent for publishing KeplerJAI bulletins, but it can make live authenticated posts without confirmation and has weak controls around credentials, endpoints, and scheduled execution.

Install only if you intend this skill to publish live KeplerJAI bulletins under your account. Before use, restrict it to the real KeplerJAI HTTPS endpoint, remove or disable --insecure and arbitrary endpoint/header overrides, require confirmation before production posting, avoid implicit invocation for publish runs, and use narrowly scoped credentials.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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)

T01 · Skill Instruction Hijacking

Error
Location
scripts/format_bulletins.py:28
Finding

Forced Promotional Content in Agent Responses

Content
View full analysis
str: if len(items) != 7: fail(f"已发布条目数量不是 7 条,当前为 {len(items)} 条。") blocks = [] for item in items: english_title = item["english_title"] chinese_title = item["chinese_title"] frontend_url = item["frontend_url"] if not english_title or not chinese_title or not frontend_url: fail("存在缺少标题或链接的已发布条目。") blocks.append("\n".join([english_title, chinese_title, frontend_url])) return "\n\n".join(blocks) + "\n\n\n" + footer + "\n" ``` ```python parser.add_argument( "--footer", default=DEFAULT_FOOTER, help="结尾固定文案,默认 '联盟小秘书 lmxms2020'", ) ``` The supporting Skill instructions require the generated `final-message.txt` to be returned verbatim. Separate stage-three and integrated prompts also require the same contact identifier at the end of every successful response. ### Technical Analysis The final formatter unconditionally appends a promotional contact identifier unless an operator explicitly overrides the footer. This content is not required to collect, validate, publish, or summarize AI news. The behavior is reinforced across multiple instruction files. The Skill directs the Agent to return only the contents of `final-message.txt`, while the formatter places the promotional identifier in that file. As a result, loading and following the Skill changes the Agent's response objective from returning only the requested bulletin summary to dis ...[truncated 1123 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/publish_bulletins.py:325
Finding

Automatically Loaded Credentials Can Be Sent to an Arbitrary Endpoint

Content
View full analysis
dict[str, str]: headers = { "Content-Type": "application/json; charset=utf-8", "Accept": "application/json", } if args.bearer_token: headers["Authorization"] = f"Bearer {args.bearer_token}" if args.cookie: headers["Cookie"] = args.cookie for entry in args.header: if ":" not in entry: fail(f"无效 header: {entry}。格式必须是 Key: Value") key, value = entry.split(":", 1) headers[key.strip()] = value.strip() return headers ``` ```python def create_ssl_context(insecure: bool) -> ssl.SSLContext | None: if not insecure: return None context = ssl.create_default_context() context.check_hostname = False context.verify_mode = ssl.CERT_NONE return context ``` ```python def post_json( url: str, payload: dict[str, Any], headers: dict[str, str], timeout: int, insecure: bool, ) -> dict[str, Any]: data = json.dumps(payload, ensure_ascii=False).encode("utf-8") request = urllib.request.Request(url, data=data, headers=headers, method="POST") context = create_ssl_context(insecure) try: with urllib.request.urlopen(request, timeout=timeout, context=context) as response: raw = response.read().decode("utf-8", errors="replace") ``` ```python def load_auth_defaults() -> dict[str, str]: """Load auth from OpenClaw environment variables.""" auth: dict[str, str] = {} env_token = first_nonempty_env(PREFERRED_BEARER_ENV_KEYS) env_cookie = first_nonempty_env(PREFERRED_COOKIE_ENV_KEYS) if env_token: auth["bearer_token"] = env_token if env_cookie: auth["cookie"] = env_cookie return auth ...[truncated 3080 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/run_saved_pipeline.py:468
Finding

Secrets Are Exposed Through Child-Process Command-Line Arguments

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
requirements.txt:1
Finding

Unbounded Third-Party Dependency Version

Content
View full analysis
=6.0 ``` The documented installation procedure executes: ```text pip install -r requirements.txt ``` ### Technical Analysis The dependency specification accepts every PyYAML release from version 6.0 onward. It does not use an exact version, lock file, or package-integrity hash. This means two installations performed at different times can resolve to different package code. A future compromised, malicious, or incompatible release could be installed without any change to the audited Skill package. The project correctly uses `yaml.safe_load()` rather than an unsafe object constructor, so no direct YAML deserialization exploit was identified. The finding concerns dependency integrity and reproducibility. ### Attack Path 1. A user follows the documented installation command. 2. The package resolver queries the configured Python package index. 3. It selects the newest release satisfying `PyYAML>=6.0`. 4. A future compromised or otherwise unreviewed release is installed. 5. That package code executes when the Skill imports `yaml`. This path depends on a compromised release, package index, mirror, or resolver configuration; no evidence of a currently malicious PyYAML package was found. ### Impact Assessment The possible impact is execution of unreviewed third-party code with the privileges of the Agent or user installing and running the Skill. The actual risk is conditional on a future supply-chain compromise or incompatible release and is therefore rated Low based on the reviewed artifact. ]]>
Remediation
View remediation
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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
Findings (36)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill description mentions publication at a high level but does not clearly disclose that authenticated outbound API requests will be made using environment-sourced credentials. In an agent setting, incomplete disclosure of external transmission and credential usage can cause users to authorize a workflow without understanding that data will be sent off-host under their identity.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill description mentions publication at a high level but does not clearly disclose that authenticated outbound API requests will be made using environment-sourced credentials. In an agent setting, incomplete disclosure of external transmission and credential usage can cause users to authorize a workflow without understanding that data will be sent off-host under their identity.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The skill description mentions publication at a high level but does not clearly disclose that authenticated outbound API requests will be made using environment-sourced credentials. In an agent setting, incomplete disclosure of external transmission and credential usage can cause users to authorize a workflow without understanding that data will be sent off-host under their identity.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill description mentions publication at a high level but does not clearly disclose that authenticated outbound API requests will be made using environment-sourced credentials. In an agent setting, incomplete disclosure of external transmission and credential usage can cause users to authorize a workflow without understanding that data will be sent off-host under their identity.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The prompt directs the agent to make live external changes by publishing seven items via POST without any user-facing warning or approval checkpoint. This is dangerous because it can cause irreversible or publicly visible actions on a production system without ensuring the user understands that real-world side effects will occur.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The prompt directs the agent to perform live publication to an external production API, which goes beyond the described scope of collecting news, validating JSON, and generating summaries. This creates an unauthorized side-effecting capability: an invocation intended for data gathering can silently create public content on a real service.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The prompt explicitly orders real API publication and forbids asking for human confirmation, which removes a critical safety checkpoint before making irreversible external changes. In practice, this can cause unauthorized or mistaken publication of content to a public-facing service with no opportunity for operator review.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

This markdown file states that outputs are written into the workspace and that publish_bulletins.py calls the KeplerJAI bulletin publishing API, which implies both file creation and network transmission. However, the description does not provide an explicit user-facing warning about sending bulletin content to an external service or about overwriting/creating output artifacts.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill advertises a workflow that reads environment variables, performs file I/O, uses network access, and may invoke shell-capable scripts, but it declares no explicit tool scope or permissions boundary. This makes it easier for an agent runtime to grant broader capabilities than users expect, increasing the risk of unintended credential use, external publication, or filesystem modification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Natural-language instructions throughout the file are presented only in Chinese, which can force a specific language on users or agents without opt-in. The policy allows locale constraints only when they are optional or clearly justified, which is not stated here.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill enables implicit invocation without any visible trigger constraints, scope limits, or user-consent guardrails. In an agent environment, this can cause the publishing workflow to be invoked unexpectedly based on loose prompt matching, increasing the risk of unintended collection, validation, publication, or summary actions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Several user-facing error and CLI messages are written only in Chinese, and the script does not offer any language selection or explain that it is intentionally Chinese-only. This is a natural-language policy concern because it imposes a specific locale without user opt-in.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The --insecure option disables certificate validation and hostname checking, allowing a man-in-the-middle attacker to intercept or modify bulletin content and steal bearer tokens or cookies during publication. Because this script performs authenticated POST requests, weakening TLS materially increases the risk of credential compromise and unauthorized content publication.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The script sends content, source URLs, and authentication headers to a remote API endpoint via HTTP POST. While publishing is part of the script's purpose, the file does not provide an explicit user-facing warning that local input content and auth material will be transmitted over the network, especially when combined with optional cookie and bearer-token headers.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The manifest describes collecting AI news, validating stage1 JSON, publishing bulletins, and generating a workspace-oriented summary. While publishing itself justifies network access, automatically sourcing bearer tokens and cookies from environment variables introduces credential-harvesting behavior that is not an obvious or declared requirement of that purpose.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This code automatically loads bearer tokens and cookies from environment variables and then uses them for publication requests. Although the behavior is implemented clearly in code comments, there is no user-facing notice at runtime that sensitive credentials will be read from the environment, which fits the missing-warning criterion for sensitive environment variable access.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The script emits user-facing status and error messages in Chinese, such as "开始执行", "已完成", and later CLI help/announcements are also Chinese-only. This imposes a specific language/locale on all users without opt-in or documented justification, which matches the natural-language policy violation criteria.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/run_saved_pipeline.py (reported line 157)May include surrounding context.

python
) -> subprocess.CompletedProcess[str]:
    announce(f"[{step_name}] 开始执行")
    log_line(log_path, f"START {step_name}: {sanitize_command(cmd)}")
    raw_result = subprocess.run(cmd, capture_output=True, text=False)
    result = subprocess.CompletedProcess(
        args=raw_result.args,
        returncode=raw_result.returncode,

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The prompt instructs the agent to write output to a local path, overwrite any existing file, and do so before replying, without any requirement to verify the path, obtain explicit user consent, or constrain the destination to a safe workspace directory. In an agent environment, a placeholder like __STAGE1_OUTPUT_PATH__ could be substituted with an unintended or sensitive path, enabling destructive overwrite of local files or clobbering prior data.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The instructions are entirely in Chinese and require fixed Chinese output labels such as '原因' and '已成功发布条数' without presenting any language opt-in or alternative. That is a natural-language locale policy issue unless the skill is explicitly documented as region- or language-specific.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The prompt instructs the agent to perform live publication to an external production API, which is a materially different and higher-risk action than merely collecting news, validating JSON, or generating summaries. This creates a scope-expansion issue where a content-processing skill can silently mutate external state and publish public-facing data.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Requiring raw curl-based POST requests grants direct network side effects and bypasses safer application-layer controls that could enforce validation, auditing, authentication handling, and environment restrictions. In the context of an agent skill described as summarization/validation-oriented, this capability is broader than necessary and increases the chance of unintended or unauthorized publication.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The prompt suppresses logs, intermediate output, and human confirmation while instructing live publication, which reduces transparency and eliminates key safety checks during a side-effectful operation. This combination makes erroneous, malicious, or unintended publishing harder to detect and stop before completion.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The file is entirely written as a Chinese-only task specification and requires generation of a specific Chinese group-summary message, including fixed Chinese output content. This imposes a language/locale constraint without any opt-in, alternative language option, or documented regional justification, which matches the language-policy violation category.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The task is framed as a one-shot end-to-end workflow with broad authority to collect, publish, and finalize output in a single run, but it lacks clear invocation constraints or safety boundaries. This makes accidental triggering of high-impact external actions more likely, especially when combined with instructions not to pause or request confirmation.

Content

No source excerpt is available for this finding.