| name | suggesting-workflows |
| description | Suggests Claude Code features and workflows that would improve how you're working right now, with exact commands to use. Use when something feels inefficient, when doing repetitive manual work, when starting a new project, or when you want to discover features you didn't know existed. |
| user-invocable | false |
Suggesting Workflows
Overview
Most Claude Code users utilize about 20% of its capabilities. This skill observes what you're doing and suggests specific features that would help — with exact commands, not vague pointers.
When to Invoke
- User types
/suggest
- User says "is there a better way to do this?"
- When the user is clearly doing something manually that Claude Code can automate
- When starting a new type of project or workflow
How It Works
Step 1: Observe Current Workflow
Analyze the conversation context for patterns:
- What task is the user working on?
- What tools and commands have been used?
- What's being done manually that could be automated?
- What scale is the work (single file? multi-file? cross-repo?)
Step 2: Cross-Reference Features
Read the feature catalog at ${CLAUDE_SKILL_DIR}/feature-catalog.md for a comprehensive map of Claude Code capabilities organized by user intent.
If claude-docs-skill is installed, read relevant documentation sections for detailed feature information and current syntax.
Step 3: Output 1-3 Suggestions
For each suggestion:
Suggestion:
What you're doing: <describe the current manual/suboptimal approach>
Better approach:
How to use it:
<exact command, config, or setup — copy-pasteable>
Why it's better:
Step 4: Skip What's Already Working
Don't suggest features the user is already using effectively. If they're already using /batch for parallel work, don't suggest /batch.
Common Suggestion Patterns
| What You're Doing | Suggest |
|---|
| Running tests manually after each change | /loop 30s pnpm test — auto-reruns on interval |
| Implementing 5+ independent files sequentially | /batch <plan> — parallel worktree agents |
| Manually checking if work is done | verification-before-completion skill (if TaskPlex installed) |
| Repeating the same instructions every session | Create a CLAUDE.md or project-level .claude/rules/*.md |
| Copy-pasting between sessions | Auto-memory system — Claude remembers across sessions |
| Not sure what approach to take | /plan — enter plan mode for design-first thinking |
| Debugging by trial and error | systematic-debugging skill (if TaskPlex installed) |
| Writing code before tests | test-driven-development skill (if TaskPlex installed) |
| Want to automate a recurring check | Hooks — PostToolUse, UserPromptSubmit, Stop events |
| Working on multiple features | Git worktrees — isolated branches without switching |
| Need external service data | MCP servers — connect to APIs, databases, services |
| Large codebase, slow navigation | Explore agent type — fast codebase exploration |
| Repetitive prompt patterns | Skills — create reusable .claude/skills/ markdown files |
What Makes a Good Suggestion
- Specific, not vague. Not "you could use hooks" but "add a PostToolUse hook that runs
pnpm typecheck after every Edit — here's the hooks.json entry."
- Copy-pasteable. The "How to use it" section should be something the user can run immediately.
- Relevant to NOW. Suggest features for the current task, not hypothetical future needs.
- Honest about tradeoffs. If a feature has setup cost, say so. "This takes 5 minutes to set up but saves you from manually running tests every time."
Edge Cases
- User is already optimal: Say so. "Your current workflow is solid — no suggestions this time."
- claude-docs-skill not installed: Fall back to feature-catalog.md. Mention that installing claude-docs-skill would give more detailed, current feature information.
- Multiple equally good suggestions: Prioritize by impact — which one saves the most time or prevents the most errors?
