claude-leverage

Slash commands are `.md` files that define reusable prompts invoked with `/<command-name>` in Claude Code.

Delegate code review to the code-reviewer subagent (Sonnet, read-only) and apply approved fixes in the main session

Smart commit - main session handles trivial commits directly, delegates non-trivial ones to git-committer subagent (Sonnet) to save Opus context

Gather implementation context before starting a task. Delegates to context-gatherer subagent (Sonnet) to pre-fetch everything Opus needs.

Delegate test execution to the test-runner subagent (Sonnet) and apply approved fixes in the main session

Subagents are specialized agents spawned by your main Claude Code session. Each runs with its own context window, a restricted set of tools, and an explicitly declared model. The main session delegates work to them and receives a structured result back.

Use after writing or modifying code. Use proactively before commits or PRs. Use when user asks for review, audit, or feedback on code. Read-only - returns structured feedback, never modifies files.

Use before implementing features or fixing bugs that touch multiple files. Gathers all relevant context (types, interfaces, patterns, dependencies, tests) into a structured package so Opus can implement without exploring itself. Read-only — returns structured context, does not propose solutions or write code.

Use ONLY for trivial commits explicitly marked as such (single-file change, under 20 lines, no security-sensitive paths). For anything beyond that, use git-committer (Sonnet) instead. Optimized for speed and cost on pure plumbing tasks.

Stage, commit, and push non-trivial changes using Conventional Commits. Reads diff, matches repo style, reports result. Does not modify code. For trivial single-file commits under 20 lines, see git-committer-quick (Haiku) instead.

Use for codebase discovery. Use to find where something is defined or imported. Use to identify patterns used across files. Use proactively before making changes that might affect multiple files. Read-only - returns structured findings report. Does not propose changes, suggest refactoring, or write code.

Use to understand how something works in the codebase before making changes. Use when asked to explain a pattern or flow that spans multiple files. Use proactively when planning non-trivial changes that depend on existing patterns. Read-only - returns structured synthesis. Does not propose changes, suggest refactoring, or write code.

Use after writing or modifying code to verify tests pass. Use proactively before commits. Use when user asks to run tests, check for regressions, or diagnose test failures. Reports structured failure analysis - never modifies code or test files.

2 hooks across 1 event