The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the local system by reading the output of ps and launchctl commands via scripts/load_census.sh. This represents an indirect prompt injection surface as process names and command-line arguments are untrusted strings that could be crafted to include instructions. The skill lacks explicit boundary markers or sanitization for this data, though its focus on numeric aggregation (CPU time, child counts) mitigates the impact.
[COMMAND_EXECUTION]: The skill utilizes several standard macOS diagnostic commands including sysctl, uptime, ps, and launchctl. These are used for their intended purpose of system monitoring and performance profiling. It also executes a local shell script, scripts/load_census.sh, which performs safe data aggregation using standard tools like awk and sort.
[SAFE]: The skill instructions emphasize a safe operational boundary, stating that diagnosis is read-only and that any remediation (such as restarting services or terminating processes) requires the current user's explicit authorization. It specifically warns the agent not to act on other sessions or users' processes without consent.

macos-load-doctor