infisical-setup
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides installation commands for the Infisical CLI using patterns like
curl -1sLf ... | bashandwget -qO- ... | shinreferences/cli-setup.mdandreferences/docker-integration.md. These targetartifacts-cli.infisical.com, which is a vendor-owned domain belonging to the skill author (infisical). As these are official distribution channels for the intended service, they are considered safe. - [DYNAMIC_CONTEXT_INJECTION]: In
references/agent-bootstrap.md, the skill describes the use of the! infisical loginsyntax. This is used as a fallback mechanism for users to manually authenticate a shared session when an automated subprocess login fails. This use case is a documented feature of specific agent platforms for legitimate session synchronization. - [DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages highly sensitive data (secrets). However, it includes proactive security instructions in
references/agent-bootstrap.mdsuch as 'Don't open, read, or print the user's .env files' and 'Never add --output or --show-values'. It also suggests adding .env files to.gitignoreand rotating any secrets that were previously committed to version control. - [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface via the ingestion of local
.envfiles through theinfisical secrets set --filecommand. The mandatory evidence chain is as follows: - Ingestion points:
.envfiles in the project root ingested viainfisical secrets set --file. - Boundary markers: The skill relies on the Infisical CLI's internal parser; it explicitly instructs the AI agent to avoid reading the file contents directly.
- Capability inventory: The skill can write secrets to a remote vault and influence how applications are started via
infisical run. - Sanitization: The instruction set directs the user/agent to check for specific formatting errors (empty values, spaces in keys) rather than executing content from the files. Given the security-focused instructions and vendor-controlled tools, this surface is considered low risk.
Audit Metadata