web-to-markdown
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute local system commands (
command -v,npm install,web2md,mkdir -p,ls -la,wc -c) based on user-supplied options. This poses a potential command execution capability risk depending on how input parameters like URL and output path are sanitized. - [INDIRECT_PROMPT_INJECTION]: The primary function of the skill is to fetch and process untrusted external web contents via
web2mdand feed the resulting Markdown back into the agent's context. An attacker could place malicious instructions on a web page that, when processed, attempts to influence the agent's downstream behavior. - Ingestion points: External URLs fetched via
web2md(SKILL.md) - Boundary markers: None specified in instructions to contain or delimit the fetched Markdown data.
- Capability inventory: Executes shell commands (
command -v,npm install,web2md,mkdir,ls,wc) across workflow steps (SKILL.md). - Sanitization: Standard single-quoting is mentioned (
web2md '<url>'), but no strict input verification or handling of malicious content inside the generated Markdown is defined.
Audit Metadata