Claude Automation Advisor
Analyze a codebase and identify the most useful Claude Code automations for the project.
The skill declares read-only analysis and user-managed implementation, reducing direct external impact. However, it is granted Bash, encourages web search and MCP recommendations, and includes examples involving CI, deployment, and Edit/Write permissions without defining least privilege, user confirmation, sensitive-data handling, data-flow disclosure, dependency review, or rollback, so points are deducted.
The workflow is organized into codebase analysis, recommendation generation, and report output, with reference files and detection tables supporting common paths. Still, there is no test suite, execution evidence, abnormal-input handling, tool-failure diagnosis, or clear failure feedback, and some behavior depends on external search and environment-specific tools, so the static score remains below the cap.
Audience, trigger scenarios, automation categories, and report structure are reasonably clear, covering hooks, subagents, skills, plugins, and MCP. Non-fit boundaries, exclusion conditions, and parameter limits are under-specified; Chinese-language support and mainland-China reachability are not addressed, and core web-search guidance may depend on network availability, so points are deducted.
The documentation is readable and progressively organized, with installation locations, configuration examples, invocation controls, references, and detection patterns. It lacks skill-specific versioning, changelog, named maintenance responsibility, update path, FAQ, explicit limitations, and stable dependency/version notes, so points are deducted.
The skill can map codebase signals to categorized recommendations and supplies directly reusable configuration and report templates, covering the core task. However, recommendation quality depends on model interpretation and external search; representative outputs are not verified and alternatives or marginal benefit are not systematically compared, so the static cap applies.
The supplied skill, reference files, and repository-level CI descriptions are auditable primary material with concrete detection rules. However, there are no skill-specific tests or third-party execution results; the CI evidence mainly checks repository plugins and MCP URL liveness, not this skill's key behavior, so the score does not exceed the static cap.
- Despite the read-only claim, Bash can execute arbitrary commands; usage should constrain command scope and explicitly prohibit file modification, commits, or sensitive-data disclosure.
- MCP, web search, and recommended external services may involve data transfer, credentials, and network reachability; reports should disclose data flows, request user confirmation, and verify service trustworthiness.
- The supplied material does not prove that recommendations were executed or validated; users should review detection results, version compatibility, and external-service permissions manually.
What it does & when to use it
This skill inspects a project’s languages, frameworks, dependencies, tests, CI/CD configuration, and existing Claude Code setup before making recommendations. It covers hooks, subagents, skills, plugins, and MCP servers. The skill is explicitly read-only: it analyzes the codebase and produces recommendations without creating or modifying files. It normally surfaces one or two high-value recommendations per category, with broader coverage when the user requests a specific category.
Uses Read, Glob, Grep, and Bash to inspect project files, directory structures, dependency manifests, and Claude Code configuration. It runs shell commands to identify project types, frameworks, libraries, tests, and deployment-related signals. It maps those signals to recommendations for hooks, subagents, skills, plugins, and MCP servers, including reasons, locations, and selected configuration examples. Its instructions also call for web search when recommendations should reflect the codebase’s specific tools, frameworks, or libraries.
- A developer is setting up Claude Code for a new project and wants a practical starting set of extensions.
- A team using a frontend framework, backend service, database, or external API wants MCP recommendations tied to its stack.
- A developer has formatters, linters, type checks, or tests and wants suitable post-edit or protection hooks.
- An engineering team wants repeatable skills for API documentation, test generation, component creation, release notes, or project conventions.
- A maintainer needs suggestions for security reviewers, performance analyzers, UI reviewers, or parallel code-review subagents.
Pros & cons
- Covers five major Claude Code extension categories: hooks, subagents, skills, plugins, and MCP servers.
- Uses repository signals such as dependencies, tests, directories, and configuration to tailor recommendations.
- Clearly operates in read-only mode and does not directly alter project files or settings.
- A constrained one-to-two-per-category output is useful for initial setup decisions.
- Recommendation quality depends on detectable files, directories, and configuration; the source provides no test results or accuracy evidence.
- The instructions call for web search, although web search is not explicitly listed in the frontmatter tools.
- Its recommendations are centered on Claude Code mechanisms and require adaptation for other clients.
- It does not implement the recommendations, so users must create or configure them separately.
How to install
The skill is located at plugins/claude-code-setup/skills/claude-automation-recommender/SKILL.md. The README documents installation for the marketplace collection: in Claude Code, run /plugin install {plugin-name}@claude-plugins-official, or browse through /plugin > Discover. The supplied source does not document a standalone installation command for this skill or identify the exact parent plugin name.
How to use
In a Claude Code environment where the skill is installed and discoverable, ask a prompt such as “Analyze this project and recommend Claude Code automations” or “Which hooks, skills, and MCP servers should I set up for this codebase?” The skill reads the repository and returns recommendations; it does not implement them. The source does not define a more specific command name or parameter format.