Back to skill

Security audit

AI 提及与引用复测

Security checks for vulnerabilities and agentic risk

Overview

The skill’s local analysis and optional account-based collection fit its stated purpose, with privacy and dependency precautions.

Keep exports and reports private because reports retain original records. Optional online sampling sends questions to the selected services through your logged-in accounts; review sensitive content first and approve content before publishing. Use idna 3.15 or newer within the declared range.

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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • 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 (14)

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · README.en.md (reported line 77)May include surrounding context.

python3 scripts/self_test.py

text

Controlled synthetic tests verify accounting, paired changes, missing citations, all-failed runs, domain boundaries, invalid inputs, and relocation. They do not query AI services or access credentials. See [example details](examples/README.md).

The monitoring design references [GEO Optimization 1.1.0](https://clawhub.ai/capt-marbles/geo-optimization). It retains question categories, historical records, category summaries, and separate answer/citation evidence. This independent implementation adds configurable brand/domain/question inputs, complete history reports, strict paired comparisons, explicit planned/effective/citation-observed sample sizes, and offline export analysis. See [NOTICE.md](NOTICE.md) and [LICENSE](LICENSE) for provenance and licensing.

Unvalidated Output Injection

High
Category
Output Handling
Confidence
95% confidence
Finding

Model output is used without validation or sanitization. Unvalidated output injected into downstream contexts (SQL, shell, HTML) enables injection attacks and arbitrary code execution.

Content

Scanner excerpt · scripts/self_test.py (reported line 305)May include surrounding context.

python
class CliTests(unittest.TestCase):
    def run_cli(self, input_file, output, cwd, script=None, extra=()):
        return subprocess.run([sys.executable, str(script or ROOT / "scripts/citation_monitor.py"),
                               "--input", str(input_file), "--output", str(output), *extra],
                              cwd=cwd, capture_output=True, text=True, check=False)

The online collection workflow lacks a privacy warning about submitting user-provided questions to external platforms.

Medium
Category
Output Handling
Confidence
86% confidence
Finding

This paragraph instructs the agent to perform browser-based collection using an existing authenticated session, rather than merely listing an available tool. It does not warn that collection submits the user's questions and potentially sensitive brand or research context to external services under that account.

Content

No source excerpt is available for this finding.

Online sampling instructions omit a privacy warning about submitting user questions through logged-in third-party accounts.

Medium
Category
Output Handling
Confidence
87% confidence
Finding

The workflow instructs the agent to collect answers using Easy WebBridge and existing logged-in browser sessions. It does not warn that the supplied questions and context will be disclosed to external platforms and may be retained or associated with those accounts; the offline-analysis and self-test assurances do not cover this online workflow.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: 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.

Online sampling submits users’ original questions to external AI services through logged-in accounts without a privacy warning.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The optional sampling workflow directs the agent to submit users’ original questions to external AI services using existing logged-in accounts, without checking for sensitive content or explaining the disclosure boundary. Questions containing confidential business details or personal information could consequently enter provider records or account history; requiring the user to request sampling reduces, but does not eliminate, this conditional privacy risk.

Content

No source excerpt is available for this finding.

The sampling workflow sends original user questions and contextual data to detection executors without a privacy warning.

Medium
Category
Output Handling
Confidence
89% confidence
Finding

The instructions direct the agent to submit failed or unrun questions to aimeGeo's available detection executors while retaining the original questions and search/context conditions. They do not warn that this information may be transmitted to the configured model or platform services, or advise reviewing it for sensitive content before submission.

Content

No source excerpt is available for this finding.

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · examples/success/output/前后对照报告.md (reported line 8)May include surrounding context.

md
计划 12;真实有效回答 0;原生引用可判定回答 0。
提及:N/A(有效分母 0);原生域名引用:N/A(有效分母 0)。

状态:success=0;error=0;timeout=0;not_run=0;simulated=12。
真实有效回答中有 0 条未提供原生引用列表,未按零引用处理。

此输入为模拟样例;模拟命中仅保留在 JSON 的 simulation_preview,不代表平台实际提及、引用或收录。

Generated reports force Chinese without a user language choice or documented locale justification.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The report generator hardcodes Chinese headings and explanatory text regardless of the user's language or input content. The CLI provides no language selection, and this file documents no region-specific requirement that would justify enforcing Chinese; failures_text likewise produces Chinese report text.

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/self_test.py (reported line 305)May include surrounding context.

python
class CliTests(unittest.TestCase):
    def run_cli(self, input_file, output, cwd, script=None, extra=()):
        return subprocess.run([sys.executable, str(script or ROOT / "scripts/citation_monitor.py"),
                               "--input", str(input_file), "--output", str(output), *extra],
                              cwd=cwd, capture_output=True, text=True, check=False)

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · LICENSE (reported line 12)May include surrounding context.

text
permit persons to whom the Software is furnished to do so.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A
PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · NOTICE.md (reported line 37)May include surrounding context.

md
permit persons to whom the Software is furnished to do so.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A
PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE

Report labels are fixed to Chinese without a user language choice or a documented locale-specific justification.

Low
Category
Output Handling
Confidence
92% confidence
Finding

The description specifies that reports use Chinese labels, including when invoked through this English-language README. It provides neither an opt-in or language selection nor a justification requiring Chinese-language output.

Content

No source excerpt is available for this finding.

Unverifiable Dependency: idna has 4 known advisory(ies) (CVE-2026-45409 (Internationalized Domain Names in Applications (IDNA): Specially crafted inputs ); CVE-2024-3651 (Internationalized Domain Names in Applications (IDNA) vulnerable to denial of se); CVE-2024-3651 (A vulnerability was identified in the kjd/idna library, specifically within the ) +1 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
40% confidence
Finding

OSV returned advisories for this package, but its resolved version is unknown. A package-name lookup does not establish whether the installed release is affected by those advisories.

Content

No source excerpt is available for this finding.