Back to skill

Security audit

胡田 OPC导师 股权合作机制

Security checks for vulnerabilities and agentic risk

Overview

The skill is documentation-only and purpose-aligned, but it gives high-impact financial, account, tax, credit, and immutable data-sharing guidance without enough privacy, legal, or reconciliation safeguards.

Review carefully before installing or using. Treat this as business/legal template material only, not authority for an agent to move money, configure bank access, publish member records, or put source documents on-chain. Require qualified legal, tax, privacy, and finance review; store only hashes or minimal evidence on-chain; restrict sensitive records by role; and fix the payout formula before using it for real distributions.

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 (2)

T09 · Insecure Skill Coding Practices

Warning
Location
references/03-六大系统.md:392
Finding

Overbroad disclosure and immutable storage of sensitive business records

Content
View full analysis

Vulnerability Details

File Location:

  • references/03-六大系统.md:58-66
  • references/03-六大系统.md:184-192
  • references/03-六大系统.md:392-419
  • references/04-财务税务.md:21-29
  • references/04-财务税务.md:218-252
  • references/05-信用评级.md:189-199

Vulnerability Type: Excessive data exposure and unsafe immutable storage
Risk Level: Medium

Vulnerable Documentation Excerpts

The following excerpts are faithful English translations of the source text:

text
1. Select a template based on the SPV type
2. Enter contribution weights, allocation ratios, and milestones
3. Compliance review
4. Electronic signature
5. Store signature information as evidence on a blockchain
6. Generate the initial contribution weights
text
Account operating rules:
- Authorization: authorization by two people
- Large transactions: transactions of at least 50,000 require three people
- Notification: notify all members of every operation by SMS or email in real time
- Daily automatic reconciliation
- Generate a weekly statement
text
7. Blockchain evidence storage: place all nodes on-chain

Blockchain evidence storage:
Place data from the entire system on-chain.
text
Online banking configuration:
- Query permission: open to all members
- Transfer permission: authorization by two people
text
Step 7: Tax payment evidence
Generate tax payment evidence and make it available to all members.
Each member declares individual income tax independently.
text
Credit disclosure:
- Monthly credit score and grade: available to all members
- Upgraded member list: available to all members
- Downgrade reasons: visible to all members
- Serious misconduct: visible to all members
- Complete scoring details: visible only to the affected individual

Technical Analysis

The proposed workflow processes contracts, electronic signatures, invoices, bank records, ...[truncated 2600 chars]

Remediation
View remediation

Remediation Suggestions

  1. Keep source records off-chain. Store contracts, invoices, bank records, tax evidence, signatures, and disciplinary evidence in an encrypted document repository. Put only a salted cryptographic hash, timestamp, record type, and non-sensitive identifier on-chain.
  2. Implement project-scoped RBAC. Separate permissions for project members, finance operators, supervisors, compliance personnel, and auditors. Membership alone must not provide access to all financial or contractual records.
  3. Apply field-level masking. Mask account numbers, identity numbers, telephone numbers, addresses, signatures, tax identifiers, and unrelated transaction details.
  4. Restrict notifications. Send transaction alerts only to authorized account custodians, finance personnel, and designated supervisors. Other members should receive sanitized aggregate reports.
  5. Minimize reputation disclosure. Make detailed scoring evidence available only to the affected person and authorized reviewers. Publish only the minimum status required for project eligibility.
  6. Define retention and correction controls. Establish retention periods, secure deletion procedures, correction workflows, appeal handling, and methods for superseding incorrect blockchain commitments.
  7. Require encryption and audit logging. Encrypt data in transit and at rest, use strong authentication, and record access to sensitive documents in tamper-evident audit logs.
  8. Perform a privacy and data-flow review. Classify every stored field and document before implementing the proposed platform or integrating electronic-signature, payment, or blockchain providers.

T09 · Insecure Skill Coding Practices

Warning
Location
references/04-财务税务.md:107
Finding

Unnormalized credit coefficients produce incorrect revenue allocations

Content
View full analysis

Vulnerability Details

File Location:

  • references/04-财务税务.md:107-117
  • references/04-财务税务.md:121-141
  • references/05-信用评级.md:91-109

Vulnerability Type: Financial business-logic error
Risk Level: Medium

Vulnerable Documentation Excerpts

The following excerpts are faithful English translations of the source text:

text
Formula 1: Distributable amount
Distributable amount =
    project receipts - reserved taxes - platform service fee - actual costs

Formula 2: Allocation to each member
Member allocation =
    distributable amount × member contribution weight × credit coefficient

Formula 3: Final received amount
Final received amount =
    base allocation × credit coefficient
text
Assumptions:
- Project receipts: 100,000
- Confirmed cost invoices: 20,000
- Contribution weights A:B:C = 5:3:2
- Credit coefficients: A=1.0, B=0.9, C=1.1

Remaining distributable amount: 69,000

A allocation: 69,000 × 50% × 1.0 = 34,500
B allocation: 69,000 × 30% × 0.9 = 18,630
C allocation: 69,000 × 20% × 1.1 = 15,180

Total: 34,500 + 18,630 + 15,180 = 68,310

Check: 69,000 = 34,500 + 18,630 + 15,180
text
Credit coefficients:
S: 1.2
A: 1.0
B: 0.9
C: 0.7
D: 0.5

Technical Analysis

The documented verification is mathematically false:

text
34,500 + 18,630 + 15,180 = 68,310

It does not equal the stated distributable amount of 69,000. The example therefore leaves 690 unallocated.

More generally, independently multiplying contribution weights by credit coefficients does not preserve the total distributable amount. The sum of payments is:

text
distributable amount × Σ(contribution weight × credit coefficient)

That expression equals the distributable amount only when the adjusted weights happen to sum to exactly one. Because coefficients range from 0.5 to 1.2, the calculation may under-allocate or over-allocate funds.

F ...[truncated 1745 chars]

Remediation
View remediation

Remediation Suggestions

Replace independent coefficient multiplication with normalized adjusted weights:

text
raw adjusted weight_i =
    base contribution weight_i × credit coefficient_i

normalized adjusted weight_i =
    raw adjusted weight_i / Σ(raw adjusted weight)

payment_i =
    distributable amount × normalized adjusted weight_i

Apply the following additional controls:

  1. Require base contribution weights to be non-negative and sum to exactly 100% before adjustment.
  2. Reject missing, negative, non-numeric, or out-of-range coefficients.
  3. Perform monetary calculations using fixed-point decimal arithmetic rather than binary floating point.
  4. Define a deterministic rounding policy at the smallest currency unit.
  5. Assign any final rounding remainder through a documented rule, such as adding it to the largest allocation.
  6. Enforce the invariant that the sum of participant payments equals the distributable amount exactly.
  7. Separately identify taxes, reserves, platform fees, costs, and undistributed balances in the ledger.
  8. Add unit tests for all-low coefficients, all-high coefficients, mixed coefficients, one participant, zero-value projects, and currency rounding boundaries.
  9. Correct the example so its arithmetic and reconciliation are accurate.
  10. Require independent review before deploying any future allocation calculator described in scripts/README.md.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (14)

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill claims the platform is limited to tools/services and fully isolated from fund risk, but later prescribes concrete payment-routing, account governance, invoicing, tax withholding, and payout operations. This mismatch can mislead operators into treating the workflow as low-risk administrative guidance when it actually steers handling of regulated fund flows and fiduciary-like controls.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The document asserts full isolation from platform fund risk, yet later instructs detailed fund-flow governance such as customer payments to the lead party, tax reserves, service-fee deductions, reimbursements, and downstream disbursements. This contradiction can lower operator caution and encourage use of the skill for quasi-custodial or money-movement processes that carry legal, fraud, and misappropriation risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The skill handles sensitive financial, identity, contract, governance, and possibly bank-account data, but provides no user-facing notice about confidentiality, retention, sharing, legal sensitivity, or risks of putting this information into ledgers, generated files, or cross-skill transfers. That omission increases the chance of over-collection, unsafe storage, and unauthorized disclosure of highly sensitive business and personal data.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The harness validates co-managed accounts, dual-token controls, and payout procedures, moving the skill from conceptual governance into concrete financial control design. That can cause an agent to generate or normalize account-management workflows involving sensitive financial operations without adequate compliance boundaries, authorization checks, or regulated-institution safeguards.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The title and entire document are written in Chinese, and there is no indication that alternative languages are supported or that the language restriction is region-specific. Under the policy rule for natural-language violations, forcing a specific language without user opt-in is a reportable issue.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The section on external communication scripts provides mandated phrasing entirely in Chinese and frames them as normative "正确说法" without offering any language or locale choice. Under the policy, forcing a specific language without user opt-in can constitute a natural-language policy violation unless clearly justified as region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The natural-language instructions and description are entirely in Chinese, which effectively forces a specific language for users reading the skill documentation. Under the policy, language constraints should either offer a user choice or be explicitly justified as region-specific.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
76% confidence
Finding

The manifest focuses on temporary-team cooperation, compliant capability contribution, and multi-party benefit distribution. The body additionally mandates blockchain notarization ('所有节点全上链') and structured handoff outputs to other skills, which are substantive capabilities not reflected in the concise manifest description.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

All visible user-facing descriptions and operating instructions are presented in Chinese, and there is no indication that users may choose another language or that Chinese is required for a justified region-specific compliance reason. The policy requires flagging language or locale constraints when a specific language is effectively enforced without user opt-in.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The manifest describes an OPC cooperation and benefit-distribution solution, but does not declare external notarization or networked ledger interactions as part of the skill's purpose. Requiring all nodes to be 'on-chain' introduces a distinct technical capability beyond the stated business-governance scope.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The entire skill file is written exclusively in Chinese and does not indicate that other languages are supported or that Chinese is an optional user-selected locale. Under the policy rule for natural-language violations, forcing a specific language without opt-in can be a policy concern unless the locale restriction is explicitly justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

SQP-3 applies to all file types and covers language or locale policy violations. This markdown file presents all operational instructions and system descriptions only in Chinese, and nowhere indicates that users may choose another language or that the Chinese-only scope is a justified regional constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

SQP-3 applies to all file types and covers language or locale policy violations in natural-language content. This file presents all operational and compliance instructions only in Chinese and does not indicate that users may choose another language or that the locale restriction is required for a specific regional compliance purpose.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown file uses only Chinese for all headings, instructions, and checklists, which can impose a language constraint on users without any explicit opt-in or justification. Under the policy rule for natural-language violations, forcing a specific language without offering a choice is reportable unless the regional or audience limitation is clearly documented.

Content

No source excerpt is available for this finding.