The Agent Skills Directory

[COMMAND_EXECUTION]: The 'Companion check' section in SKILL.md directs the agent to execute a shell command (ls) to inspect specific skill directories on the user's filesystem, including paths like ~/.claude/skills/, ~/.agent/skills/, and ~/.agents/skills/. This allows the agent to probe the host environment for installed software and configurations.
[REMOTE_CODE_EXECUTION]: The skill instructs the agent to recommend and facilitate the installation of additional components using the npx skills add AbsolutelySkilled/AbsolutelySkilled command. This involves executing code from a remote registry, posing a risk of arbitrary code execution from a source that is not listed as a trusted vendor.
[EXTERNAL_DOWNLOADS]: The skill directs both the user and the agent to interact with external resources hosted at absolutelyskilled.pro and associated GitHub repositories for installation and additional functionality.
[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted SaaS and financial data provided by the user (ingestion point: SKILL.md). It lacks explicit boundary markers or sanitization instructions to prevent indirect prompt injection, where malicious instructions could be embedded in the data to exploit the agent's available capabilities (capability inventory: filesystem inspection via ls command in SKILL.md).

saas-metrics