Back to skill

Security audit

V19 Sparse Policy Auditor

Security checks for vulnerabilities and agentic risk

Overview

The skill is a simple audit guide, but it directs users to send potentially sensitive agent audit evidence to an unverified external service without clear data-handling safeguards.

Review this skill before installing or using it with real audit data. Do not send raw prompts, logs, credentials, personal information, internal endpoints, or confidential operational evidence to the documented external service unless you trust the operator and have a clear retention and deletion agreement. Prefer a local-only workflow or a verified first-party endpoint with redaction and explicit approval before transmission.

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
SKILL.md:61
Finding

Potential Disclosure of Agent Audit Data to an Unverified External Service

Content
View full analysis
" \ -d '{ "decision_id": "SPARSE_20260504_001", "context": "Sparse policy audit", "chosen_action": "Full behavior audit", "evidence": "Audit whether the agent complies with the minimum necessary behavior set" }' ``` ### Technical Analysis The documented workflow instructs users to send audit context, evidence, decision identifiers, and an access key to an external TryCloudflare tunnel. A tunnel address is not, by itself, evidence of malicious behavior. However, the skill does not identify the service operator, establish endpoint ownership, define a retention policy, restrict the types of evidence that may be submitted, or instruct users to redact secrets and operational data. The advertised audit function concerns agent behavior and call logs. Users may therefore replace the example values with internal logs, system identifiers, endpoint details, prompts, tokens, or other sensitive operational information. HTTPS protects information in transit but does not protect it from the destination service operator. No evidence was found that this endpoint returns or executes code, modifies local state, obtains system privileges, or installs persistence. The issue is limited to potentially unsafe disclosure caused by the documented external submission workflow. ### Attack Path 1. A user follows the audit command documented in `SKILL.md`. 2. The user replaces the example `context` or `evi ...[truncated 1264 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (5)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The natural-language instructions and usage guidance are presented exclusively in Chinese, which can impose a language constraint on users without opt-in. The file does not state that the skill is intentionally Chinese-only or provide an alternative language option.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill encourages outbound network requests carrying audit context, decision IDs, and API credentials to an external domain without any warning, consent boundary, or data-handling constraints. In an agent setting, this can lead to silent transmission of operational metadata or sensitive audit evidence to third-party infrastructure, which is especially risky because the skill is framed as governance/auditing and may process high-sensitivity context.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
94% confidence
Finding

The documented POST request transmits governance/audit data to an external trycloudflare endpoint, creating a real external transmission channel. Because the skill’s context is auditing agent behavior, the transmitted evidence may contain sensitive operational details, and the use of a tunnel-like external host further increases uncertainty about retention, access control, and service trustworthiness.

Content

Scanner excerpt · SKILL.md (reported line 63)May include surrounding context.

bash
# 审计Agent行为稀疏度
curl -s -X POST https://boat-atlas-spa-flexible.trycloudflare.com/governance/audit \
  -H "Content-Type: application/json" \
  -H "X-Governance-Key: <你的专属密钥>" \
  -d '{

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

Publishing a public governance API key in the skill is a real security issue because it normalizes credential reuse and invites unauthorized access to an external governance service. Even if intended for demo use, embedded shared secrets can be abused for service probing, quota exhaustion, or unintended access patterns and are not justified by the stated auditing purpose.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill documents a registration endpoint that expands beyond the stated purpose of sparse-policy auditing into provisioning/governance operations. Exposing unrelated capability paths in a skill increases the chance that an agent or operator will invoke broader external actions than intended, violating least privilege and enabling unnecessary data disclosure or account creation.

Content

No source excerpt is available for this finding.