The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various macOS utilities including sysctl, pmset, top, ps, and launchctl to collect performance metrics. The implementation uses Bun.spawnSync with fixed argument arrays, preventing shell command injection vulnerabilities.\n- [PRIVILEGE_ESCALATION]: High-privilege commands like powermetrics and sfltool are identified as optional and the instructions require explicit user consent before they are executed, maintaining the principle of least privilege.\n- [DATA_EXFILTRATION]: The skill sends local status notifications via curl to http://localhost:31337/notify. This activity is confined to the local loopback address and appears intended for communication with a local agent dashboard.\n- [PROMPT_INJECTION]: The skill contains an indirect prompt injection surface related to its interpretation of system-generated data.\n
Ingestion points: Output from system tools such as process names, command-line arguments, and launchd labels parsed in Tools/Vitals.ts.\n
Boundary markers: The skill does not utilize specific delimiters to isolate external system output from its diagnostic logic.\n
Capability inventory: The agent can recommend sensitive system modifications, such as killing specific process IDs or removing launch agents, based on the ingested data.\n
Sanitization: Process names and other system strings are not sanitized or escaped before being presented to the agent for analysis.\n- [PROMPT_INJECTION]: The skill includes a feature that instructs the agent to load and apply configurations from a local user directory (~/.claude/LIFEOS/USER/CUSTOMIZATIONS/SKILLS/Vitals/). These files can override the skill's default behavior, which provides a mechanism for local instruction influence.

Vitals