The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's setup instructions direct the installation of the @mhrj/mcp-agent-loop package from the NPM registry using the npx command. This package is an external dependency provided by the skill's author (meharajM).
[PROMPT_INJECTION]: The skill design creates an indirect prompt injection surface by storing and recalling untrusted data within the agent's context.
Ingestion points: Untrusted content enters the agent's memory via the init_loop tool (objective parameter), the resume_loop tool (human feedback), and through the results of actions logged via log_step which are subsequently read from the loop:// resource.
Boundary markers: The instructions do not specify the use of delimiters, XML tags, or "ignore embedded instructions" warnings when the agent reads from its persistent state file.
Capability inventory: The skill is intended for use in environments where the agent has significant capabilities, including running bash commands and editing files, which increases the risk of successful exploitation if malicious instructions are injected into the memory state.
Sanitization: No mechanism for sanitizing or validating the content stored in the persistent memory is described in the skill's instructions.

agentic-loop