The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains instructions to download a shell script from 'http://agents.allium.so/cli/install.sh' and execute it directly using the shell ('| sh'). This occurs in both the 'install' metadata field and the descriptive body of the skill. Executing unverified remote scripts is a high-risk pattern that can lead to full system compromise.
[EXTERNAL_DOWNLOADS]: The skill dynamically fetches supplementary markdown files from 'https://agents.allium.so/skills/' at runtime based on the user's request. This introduces external, unverified logic into the agent's execution environment.
[COMMAND_EXECUTION]: The skill's primary function involves installing and using a CLI tool ('allium') via shell commands, which includes network operations and payment protocol interactions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted remote data that could contain instructions designed to override agent behavior.
Ingestion points: External markdown instructions are fetched from 'https://agents.allium.so/skills/' in 'SKILL.md'.
Boundary markers: The skill does not implement delimiters or 'ignore embedded instructions' warnings for the fetched content.
Capability inventory: The skill allows for shell command execution and network requests through the 'allium' CLI and 'curl'.
Sanitization: No sanitization or validation of the remote markdown content is performed before it is integrated into the agent's context.

allium-x402