crw-self-host
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches installation scripts, GPG keys, and container images from official vendor sources including
fastcrw.com,apt.fastcrw.com, andghcr.io. - [REMOTE_CODE_EXECUTION]: Executes remote scripts and binaries as part of the installation process. Specifically:
curl -fsSL https://fastcrw.com/install | shandnpx crw-mcp. - [COMMAND_EXECUTION]: Instructs the agent to run various shell commands to install and manage the server, including
brew,cargo,docker, andpip. - [PRIVILEGE_ESCALATION]: Uses
sudoto perform administrative tasks required for system package management on Linux, such as adding APT repositories and installing system-level packages. - [INDIRECT_PROMPT_INJECTION]: The skill's functionality as a web scraper creates a potential attack surface for indirect prompt injection.
- Ingestion points: External web content is ingested through the
/v1/scrapeand/v1/searchendpoints. - Boundary markers: No explicit instructions are provided to delimit or treat scraped content as untrusted data.
- Capability inventory: The skill uses
Bashtools andReadcapabilities, which could be triggered by instructions embedded in scraped content. - Sanitization: There is no evidence of sanitization or escaping of scraped content before it is processed by the agent.
Audit Metadata