The Agent Skills Directory

[COMMAND_EXECUTION]: Modifies the '.claude/settings.json' configuration file to add a 'UserPromptSubmit' hook. This hook triggers a shell command automatically on every interaction with the agent, creating a persistent execution channel within the development environment.
[EXTERNAL_DOWNLOADS]: Fetches and executes the '@codevoyant/agent-kit' package from the npm registry via 'npx'. The package name appears to use a substitution of the author's name ('cloudvoyant' to 'codevoyant'), a technique commonly associated with typosquatting and malicious impersonation.
[REMOTE_CODE_EXECUTION]: Implements a mechanism for continuous remote code execution by hooking the agent's interaction lifecycle to an unverified external npm package.
[PROMPT_INJECTION]: Establishes an indirect prompt injection surface through its automated knowledge loading feature.
Ingestion points: 'CLAUDE.md' and 'AGENTS.md' are configured to load external data into the agent's context using shell commands.
Boundary markers: Absent; no delimiters or instructions are provided to the agent to prevent it from following commands embedded in the loaded content.
Capability inventory: The skill has the capability to write local files and configure environmental hooks for arbitrary command execution.
Sanitization: No sanitization or validation of the loaded knowledge content is performed before it enters the session context.

mem-init