Back to skill

Security audit

YoudaoNote OKF Wiki

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent YoudaoNote knowledge-base manager, but it asks agents to perform persistent cloud-note writes and local file writes with some under-scoped handling of credentials, temporary files, and automatic registry repair.

Install only if you are comfortable granting the agent read/write access to your YoudaoNote account and having it maintain OKF metadata automatically. Prefer entering API keys through a secure prompt if the CLI supports it, avoid using this on shared machines until temporary-file handling is improved, and review export paths before allowing local writes.

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
SKILL.md:59
Finding

API Key Exposed Through Command-Line Arguments

Content
View full analysis
`. ``` The original instruction explicitly requires the API key to be substituted into the command-line argument. ### Technical Analysis Passing a secret directly as a command-line argument can expose it through several operating-system and execution-layer channels: - Interactive shell history files. - Process-listing utilities while the command is running. - Terminal session recordings. - Agent tool-call logs and execution traces. - CI/CD, debugging, or command-auditing systems that record complete arguments. - Error reports that capture the invoked command. Although the key is needed to configure the legitimate YoudaoNote client, placing it directly in the argument vector does not provide adequate protection for credential material. The documentation also does not require redaction, secure prompting, log suppression, or subsequent credential rotation. ### Attack Path 1. A user follows the Skill instruction and runs `youdaonote config set apiKey `. 2. The plaintext key is stored in shell history, captured in an Agent execution trace, or temporarily exposed through the process argument list. 3. An attacker with access to the same host, execution logs, terminal history, or monitoring system retrieves the key. 4. The attacker configures a YoudaoNote client or directly accesses supported APIs with the stolen credential. 5. The attacker can access or modify cloud-note data to the extent permitted by that API key. ### Impact Assessment Successful exploitation can compromise the confidentiality and integrity of the vic ...[truncated 329 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:822
Finding

Predictable Shared Temporary Files Used for Sensitive Note Content

Content
View full analysis
", "contentFile": "/tmp/okf-xxx.md" }' | youdaonote -s okf save --json # Update an existing note youdaonote -s okf update --file /tmp/okf-xxx.md ``` The Skill repeatedly uses fixed or predictable paths, including: ```text /tmp/okf-registry.md /tmp/okf-schema.md /tmp/okf-index.md /tmp/okf-log.md /tmp/okf-raw.md /tmp/okf-page.md /tmp/okf-archive.md /tmp/okf-updated.md ``` ### Technical Analysis The files may contain private notes, imported webpages, knowledge-base metadata, conversation archives, and content intended for upload to the user’s cloud account. Predictable names under a shared temporary directory create several security risks when the host is shared with other users or untrusted processes: - Another local process may read residual content if permissions are inherited from an unsafe default umask. - An attacker may pre-create a symbolic link at an expected path and redirect the write to another accessible file. - An attacker may replace or modify the temporary file between creation and the subsequent `save` or `update` command. - Concurrent Skill executions may overwrite each other’s temporary files or upload content from the wrong session. - Sensitive files may remain on disk because the instructions do not mandate cleanup. - Falling back to a predictable relative file such as `./okf-temp.md` does not eliminate substitution or residual-data risks. The two-step write-and-upload process creates a time-of-check/time-of-use ...[truncated 2070 chars]
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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 (6)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest description is entirely written as a Chinese-only invocation and usage specification, with no indication that the skill supports user language choice or adapts to the user's locale. This can amount to a language policy violation because the skill effectively imposes a specific language without opt-in or documented justification.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 25)May include surrounding context.

md
# YoudaoNote OKF Wiki — 有道云笔记 OKF 知识库

在有道云笔记中构建符合 [OKF (Open Knowledge Format) v0.1](https://github.com/GoogleCloudPlatform/knowledge-catalog/blob/main/okf/SPEC.md) 规范的持久化知识库。
基于 OKF v0.1 规范 + [youdaonote-llm-wiki](../../skills/youdaonote-llm-wiki/SKILL.md) 成熟模式。

OKF 是一种开放的、对人类和 Agent 都友好的知识表示格式:一组 markdown 文件 + YAML frontmatter,无 schema 注册中心、无中央权威、无必需工具链。它被设计为**可读**(人类无需工具即可阅读)、**可解析**(Agent 无需专用 SDK)、**可 diff**(版本控制友好)、**可移植**(跨工具、跨组织、跨时间)。

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill states that if the registry is missing, it should be automatically rebuilt and saved 'without user awareness.' Silent persistence is risky because it causes account data mutation outside an explicit action boundary, undermining informed consent and making unintended writes harder for users to detect or audit.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill explicitly authorizes fetching arbitrary user-supplied URLs and then ingesting their contents into the knowledge base. That expands the skill from note management into unbounded external content retrieval, creating SSRF-style exposure to internal or sensitive endpoints, as well as ingestion of hostile content that can trigger downstream prompt injection or data poisoning.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The export workflow instructs the agent to create arbitrary local directories and write files to a user-provided path. This broadens the skill from cloud note management to local filesystem mutation, which can overwrite sensitive locations, plant files in unexpected paths, or be abused for persistence if the path is not tightly constrained.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown file presents the specification summary entirely in Chinese, including the introductory statement and all section content, without indicating that the language is optional or limited to a Chinese-speaking context. Under the policy rule, forcing a specific language without user opt-in is a natural-language locale violation.

Content

No source excerpt is available for this finding.