Computer Health Check
Safely diagnose your computer's health and get a scored report with actionable cleanup tips.
The skill explicitly commits to read-only operation, no sudo, cleanup only with explicit user approval, and only reversible actions. Data flow is limited to local system metadata, not sent off-machine. The repository has an MIT license but the publisher identity is unverified, and the security model is well-described (SECURITY.md makes clear these are guardrails, not a sandbox). Deductions: the underlying MCP server has powerful capabilities like terminal control, and the skill cannot enforce its own safety contract; there is no concrete implementation of a user confirmation mechanism.
The skill provides concrete command scripts for multiple platforms, but lacks error-handling guidance or output parsing details. Some commands may fail due to permissions or missing tools, but no explicit degradation strategy is provided. Static review cannot verify script execution, and there are no execution tests. Deductions: no test evidence, thin error handling, static review cannot confirm actual operation.
The skill clearly targets computer health checks with clear trigger scenarios across multiple platforms. It declares capability boundaries (read-only, no sudo) and non-fit ranges (requires admin). Environment fit is considered with commands for major OSes, but no discussion of Chinese network reachability. Deductions: no mention of external services that may be unreachable from mainland China, and no Chinese-language support mentioned.
The skill document is well-structured with version, audience, description, workflow, scoring rubric, cleanup suggestions, and pitfalls. It uses a layered design with reference files per platform. However, it lacks installation/dependency notes, troubleshooting, version history, changelog, and clear maintenance responsibility. Deductions: missing update path and changelog, insufficient installation/dependency documentation, unclear maintenance.
The skill claims to provide a comprehensive health check with recommendations, but static review cannot verify actual outcomes. It defines output format and scoring, but no real output examples are provided. Marginal value over manual execution may be limited because commands and interpretation still require user or agent effort. Deductions: no verifiable outputs or test evidence, static review cannot confirm result correctness and completeness.
The skill provides command scripts but no test cases or CI integration to verify its behavior. Documentation claims such as 'Verified on macOS 26 / Apple Silicon' are author assertions without independent evidence. The repository contains test files but none cover this skill. Deductions: lack of tests specific to the skill and third-party verification; static review only confirms documentation existence, not independent reproduction.
- The underlying Desktop Commander MCP server has powerful permissions like terminal control; its security model is guardrails not a sandbox. Please ensure the runtime environment is isolated (e.g., Docker).
- Cleanup operations require user approval, but the skill itself cannot enforce this; ensure the agent strictly follows the safety contract when in use.
- When used in mainland China, some optional checks (e.g., internet speed test) may be unavailable due to network environment.
- Static review did not verify the actual execution of scripts; test in an isolated environment before production use.
What does this skill do, and when should you use it?
This skill runs a comprehensive, read-only health check on your computer, covering disk space, memory, battery health, startup items, and resource usage. It produces a scored chat summary with prioritized recommendations, strictly adhering to a safety contract: read-only by default, no sudo, and cleanups only after explicit approval. Works on macOS, Windows, and Linux.
Runs a batch of read-only diagnostic commands through a local shell, detects the OS, reads the matching reference file, collects data on disk space, SMART health, memory pressure, battery capacity, uptime, pending updates, startup items, and current CPU/memory usage. It scores each category (🟢/🟡/🔴), computes a percentage score and verdict, outputs a structured report in chat, and suggests cleanup commands that are executed only after user approval.
- When a user says 'my computer feels slow' or 'why is my fan so loud,' quickly identify resource hogs and potential issues.
- When a user wants to 'free up disk space,' identify large reclaimable files, caches, and trash, and suggest cleanups.
- When a user wants to 'check battery health,' report cycle count and maximum capacity, and explain wear.
- When a user asks for a 'system tune-up' or 'checkup,' provide a comprehensive scored health report.
What are this skill's strengths and limitations?
- Strictly read-only by default, safe to run.
- Cross-platform (macOS, Windows, Linux) with a uniform report format.
- Provides structured scoring and clear action items, easy for users to understand.
- Cleanups are opt-in and favor reversible, non-destructive actions.
- Requires the Desktop Commander MCP to function; not standalone.
- Battery check is laptop-only and skipped on desktops/VMs.
- Relies on reference files for concrete commands, which are not included in the skill text, needing manual lookup.
- No mention of automated tests; reliability depends on script quality.
How do you install this skill?
Install the Desktop Commander MCP (e.g., via npx setup or manual config). The skill files live in plugins/claude/skills/computer-health-check/ in the repository and are typically distributed with the bundle.
How do you use this skill?
Ask your assistant directly, e.g., 'check my computer's health' or 'why is my laptop so laggy?' and it will run the diagnostics and output a report.