T09 · Insecure Skill Coding Practices
- Location
research_logger.sh:92- Finding
Command Injection Through Dynamically Constructed sed Expressions
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill has a coherent research-note purpose, but its script handles web, template, and local file data in ways that create meaningful user-safety risks.
Do not install this version for routine use until it escapes template replacements safely, validates fetched URLs, constrains template paths, uses mktemp or private temp storage, fixes the Bear fallback control flow, pins grizzly, and declares all required tools.
research_logger.sh:92Command Injection Through Dynamically Constructed sed Expressions
research_logger.sh:34Server-Side Request Forgery Through Automatic Fetching of Search Results
research_logger.sh:63Workspace Escape and Arbitrary Local File Reading Through Template Path Traversal
research_logger.sh:107Predictable Temporary-File Write Enables Symlink Clobbering and Unintended Data Persistence
SKILL.md:10Unpinned Installation of an Executable Third-Party Dependency
The script performs an additional network/content-retrieval step with web_fetch on the top search result and uses that content to generate a summary, but the declared skill behavior only mentions web search, GIF matching, and note logging. This hidden expansion of data access increases the skill’s effective permissions and can expose users to unanticipated remote content ingestion, privacy issues, or unsafe downstream processing assumptions.
The skill documentation describes automated web search, content fetching, template filling, and Bear note creation, but it does not clearly warn users that untrusted web content and template data will be written into a local Bear note. This can lead to surprising side effects, accidental persistence of misleading or sensitive content, and reduced informed consent for users running the skill.