T03 · Remote Payload Retrieval and Execution
- Location
SKILL.md:15- Finding
Unverified Remote Installer Piped Directly into a Shell
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This PowerPoint-generation skill is coherent, but it asks agents to run an unverified remote installer and execute user deck files with broad local file access.
Review before installing. Prefer installing Deno yourself through a trusted package manager, avoid the automatic remote shell installer, and only generate decks from TSX files you wrote or trust. Run the generator with path-scoped read/write permissions where possible.
SKILL.md:15Unverified Remote Installer Piped Directly into a Shell
scripts/generate.ts:33Untrusted TypeScript Deck Files Execute with Broad Filesystem Permissions
Direct external script fetching and piping to sh is a classic arbitrary code execution pattern. If the remote endpoint, transport, or upstream distribution is compromised, the host running the skill can be fully compromised during installation.
{
"id": "deno-install",
"kind": "shell",
"command": "curl -fsSL https://deno.land/install.sh | sh",
"bins": ["deno"],
"label": "Install Deno (https://deno.land)",
},
The description says to use the skill whenever a user asks to create presentations, slide decks, pitch decks, reports, or any PPTX file, which is broad enough to trigger in many common contexts. Over-broad routing can cause unnecessary activation of a skill that includes code execution and installation behavior, increasing exposure to its riskier capabilities.
The skill embeds an install step that executes a remotely fetched shell script via curl | sh. That gives arbitrary code execution on the host during skill setup, and while installing Deno is relevant to the skill, using an unchecked remote script is far more privileged and risky than necessary for presentation generation.
The installation instructions include curl -fsSL https://deno.land/install.sh | sh without any user-facing warning that it modifies the system and executes remote code. In an agent context, this is dangerous because it can normalize or automate a high-risk action without informed user approval.