agent-lang
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a DSL for processing instructions from external Markdown files, which creates an inherent surface for indirect prompt injection if those files contain malicious content.
- Ingestion points: Markdown files containing makefile blocks (e.g., SKILL.md, references/example).
- Boundary markers: The skill uses makefile code fences and specific keywords like INVARIANT and HUMAN APPROVAL REQUIRED to delimit instructions.
- Capability inventory: The skill permits the agent to execute shell commands (prefixed with @) and invoke other tools via the SLOT mechanism.
- Sanitization: It includes a mandatory HUMAN APPROVAL REQUIRED protocol and stipulates that all actions remain subject to user permissions and confirmations.
- [EXTERNAL_DOWNLOADS]: The plan and examples mention using npx langlint and npm test. These tools download and manage packages from the npm registry, a well-known service.
- [COMMAND_EXECUTION]: The skill's documentation and examples describe the execution of shell commands such as git and npm as part of its intended workflow for authoring and reviewing code.
Audit Metadata