Back to skill

Security audit

audit-budget-performance

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent budget-performance workflow, but it needs review because sensitive audit data flows through unpinned external skills and missing runtime scripts.

Review this before installing in any real audit environment. Confirm exactly which versions and publishers of the referenced helper skills and runtime scripts will run, restrict their permissions, and avoid passing confidential financial or survey data until dependency provenance and data handling are clear.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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)

T08 · Insecure Dependencies

Warning
Location
workflow.yaml:57
Finding

Unpinned External Skill Dependencies Create a Supply-Chain Risk

Content
View full analysis

Vulnerability Details

File Location: workflow.yaml:57-116
Vulnerability Type: Unpinned and externally resolved skill dependencies
Risk Level: Medium

Vulnerable Code

yaml
- id: completion_rate
  name: 绩效完成率分析
  action: data_analysis
  description: 绩效指标完成率计算和分析
  prompt: |
    分析"{project_name}"的绩效指标完成情况:
    1. 产出指标完成率(数量、质量、时效)
    2. 效益指标达成度(经济、社会、生态)
    3. 满意度指标得分
    4. 未完成指标的原因分析
  skills: [data-analyst-cn]
  depends_on: [data_collection]
  checkpoint: false
  output_key: completion_analysis

- id: budget_efficiency
  name: 资金使用效率分析
  action: data_analysis
  description: 预算资金使用效率和效益分析
  prompt: |
    分析"{project_name}"的资金使用效率:
    1. 预算执行率
    2. 资金到位率
    3. 资金节约率
    4. 成本效益分析
    5. 资金使用合规性
  skills: [data-analyst-cn, financial-fraud-detection]
  depends_on: [data_collection]
  checkpoint: false
  output_key: efficiency_analysis

- id: satisfaction_stats
  name: 满意度统计分析
  action: invoke_skill
  description: 服务对象满意度调查统计分析
  prompt: |
    对"{project_name}"的满意度调查进行统计分析:
    1. 样本代表性和有效性
    2. 满意度总体得分和分布
    3. 各维度满意度对比
    4. 不满意原因归类分析
  skills: [statistical-analysis]
  depends_on: [data_collection]
  checkpoint: false
  output_key: satisfaction_result

- id: comprehensive_scoring
  name: 综合评分
  action: invoke_skill
  description: 多维度综合绩效评分
  prompt: |
    对"{project_name}"进行综合绩效评分:
    1. 按决策、过程、产出、效益四维度打分
    2. 各指标权重和得分明细
    3. 总分和等次判定
    4. 各维度得分分析
  skills: [budget-performance-methodology]
  depends_on: [completion_analysis, budget_efficiency, satisfaction_stats]
  checkpoint: true
  output_key: final_score

- id: generate_report
  name: 编制绩效评价报告
  action: invoke_skill
  description: 生成绩效评价报告
  prompt: |
    生成"{project_name}"的绩效评价报告:
    1. 项目基本情况
    2. 绩效评价工作开展情况
    3. 绩效评价指标分析
    4. 评价结论和等次
    5. 主要问题和建议
  skills: [gov-audit-report-writer, gov-doc-formatting]
  depends_
...[truncated 2636 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin every external skill to an immutable version and, where supported, a cryptographic content digest.
  2. Resolve dependencies only from an authenticated, allowlisted registry and verify publisher identity or signatures before execution.
  3. Maintain a lock file or manifest recording the exact version, source, and digest of every resolved skill.
  4. Vendor security-critical skill definitions into the reviewed deployment artifact when feasible.
  5. Review dependency contents and transitive dependencies before approval, and repeat the review for every update.
  6. Run invoked skills with least privilege. Deny network access, arbitrary command execution, unrestricted file access, and credential access unless a specific stage requires them.
  7. Pass each skill only the minimum required data. Redact personal, financial, and confidential fields before invocation where possible.
  8. Require explicit approval for dependency changes and fail closed if the pinned implementation cannot be verified.
  9. Add runtime provenance logging so each execution records the resolved dependency version, digest, publisher, and permissions.
  10. Security-review the separately declared workflow-engine scripts before deployment because pipeline_engine.py, audit_workflow.py, and audit_router.py were referenced but not included in the audited artifact.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file presents the skill entirely in Chinese and provides the activation phrase only in Chinese, with no indication that users may interact in other languages or that the skill is intentionally limited to a Chinese-only regulatory or regional context. This creates a natural-language locale constraint without user opt-in, which falls under the language/locale policy violation category.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The skill metadata presents the name, description, tags, category, and author entirely in Chinese, with no indication that language selection is optional or that the skill is specifically intended only for a Chinese-speaking or region-specific environment. This can violate language/locale policy when users are not offered a choice or explicit notice of the enforced locale.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

The manifest hard-codes a Chinese-language analyst skill via "data-analyst-cn", which suggests a fixed language or locale behavior. Because the file does not state that the workflow is region-specific or that users can choose language, this may violate language/locale policy requirements.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

This YAML manifest describes the skill only as covering budget performance goal setting, evaluation, monitoring, and project performance assessment, but it does not specify when the workflow should or should not be invoked. In a manifest file, the absence of narrow activation scope or negative examples can make invocation boundaries ambiguous and increase the chance of unintended use.

Content

No source excerpt is available for this finding.