The Agent Skills Directory

[SAFE]: The skill is a legitimate health-check utility for the OrchestKit plugin ecosystem. Its primary function is to diagnose configuration errors, version mismatches, and broken references.
[COMMAND_EXECUTION]: The skill uses local helper scripts (Bash, Python, and Node.js) to audit the filesystem. These scripts perform safe operations such as parsing JSON files (.mcp.json, settings.json), checking file existence, and counting lines in memory files.
[DATA_EXPOSURE]: The check-agent-memory.sh script scans for sensitive patterns (AWS keys, GitHub tokens, passwords, SSNs) within agent memory directories. This is implemented as a security feature to warn the user about accidental data leaks, rather than for exfiltration.
[EXTERNAL_DOWNLOADS]: The skill references well-known external tools such as agent-browser (from Vercel Labs) and peer skills from established marketplaces. It checks for their presence and versions to ensure environment compatibility but does not perform autonomous remote code execution.
[FALSE_POSITIVE]: Static analysis flagged potential data exfiltration in references/settings-posture.md. Review confirms this content is documentation that describes how an attacker could use curl to bypass certain guards, serving as a justification for enabling the Claude Code sandbox; the skill does not execute these commands.
[SYSTEM_AUDIT]: The check-sandbox-violations.sh script queries the macOS unified log for 'Sandbox' denials. This is a read-only operation used to inform the user of background security events related to their session.

doctor