softaworks-agent-toolkit: 43 Everyday Skills for Claude Code

Most skill packs specialize; softaworks/agent-toolkit is deliberately a grab bag — 43 skills, 6 sub-agents, and 7 slash commands covering the whole shape of a working week: development, documentation, planning, testing, even workplace communication. It follows the open Agent Skills format, and every skill, agent, and command is an individually installable plugin.

Why This Skill Matters

The catalog spans ten categories. AI tools let Claude delegate to other models — codex for advanced code analysis with GPT-5.2, gemini for large-scale review across 200k+ context, perplexity for web research. Meta skills improve the toolkit itself: skill-judge evaluates skill design quality, plugin-forge builds plugins, agent-md-refactor slims down bloated instruction files.

The rest is the daily grind, handled: session-handoff for seamless AI session transfers, c4-architecture and mermaid-diagrams for documentation, gepetto for detailed implementation planning, commit-work for high-quality git commits, and — the one nobody admits wanting — humanizer, which removes AI writing patterns from text. Six specialized sub-agents (including codebase-pattern-finder and ui-ux-designer) and seven slash commands round out the set.

Installation

The quick install works across Claude Code, Codex, Cursor, and other compatible agents:

npx skills add softaworks/agent-toolkit

Claude Code users can also add it as a plugin marketplace and install piece by piece:

/plugin marketplace add softaworks/agent-toolkit
/plugin install skill-judge@agent-toolkit

Updates run through the marketplace's Update option, with an auto-update toggle if you want new versions automatically. Manual installation is a plain copy into ~/.claude/skills/.

Real Workflow: An End-of-Session Handoff

Context loss between sessions is where agent work quietly decays. Run session-handoff at the end of a working session and it produces the transfer document — what was done, what is in flight, what breaks if you stop — so the next session starts with the state instead of re-deriving it. Pair it with lesson-learned, which extracts software-engineering lessons from the session's recent code changes, and /explain-pr-changes when the work is ready for review.

Tips

• Install the three or four skills matching this week's work; each is an individual plugin, so the toolkit never has to arrive all at once.
• The agents and commands are Claude Code-exclusive per the README — other agents get the skills only.
• /codex-plan creates implementation plans with Codex 5.2 before you start coding.
• For claude.ai users, the README documents adding skills to project knowledge or pasting SKILL.md contents into the conversation.

When Not to Use This

A grab bag has no center of gravity — if you only need deep expertise in one domain (security scanning, design systems, web scraping), a specialized pack goes deeper than these generalist skills. And the professional category (standup updates, difficult-conversation coaching) presumes you want an AI in those corners of your work life; plenty of engineers sensibly do not.

See the leaderboard for more skills.