Whip — Full-Stack Development Workflow Skills

A curated AI coding-agent skills collection for the solo full-stack developer.

English | 中文

---

Whip is a hand-curated set of AI coding-agent skills built for the full-stack developer running a one-person company. It distills the best methodology from three popular skills repositories — mattpocock/skills, obra/superpowers, and addyosmani/agent-skills — then deeply rewrites them for a lean, solo workflow: less ceremony, same discipline.

All skills are fully rewritten (not copied) in English, sized for real daily use, and follow the Agent Skills specification.

Why Whip

The three source repositories are excellent but carry heavy assumptions — GitHub-Issues-driven workflows, team code review loops, enterprise observability stacks. A solo developer doesn't need most of that. Whip keeps what actually prevents mistakes and drops what slows you down:

• Tiered process — small tasks run a light loop (tdd → verify → git); only new features, cross-file changes, and releases add the heavier stages.
• Evidence before claims — a hard rule that you never say "done" without fresh verification output.
• Full-stack coverage — API design and security as always-on reference skills, not afterthoughts.
• No subagent tool? No problem — delegate spawns isolated agents via pi -p when your harness lacks native subagents.

The Pipeline

  small task (field tweak / style fix / self-contained change):
      tdd → verify → git
  medium task (new feature / cross-file change):
      design → tdd → verify → git
  large task (new project / architecture / release):
      design → tdd → verify → git → ship

Skills

Skill	What it does	When to use
design	Brainstorm mode (one-question-at-a-time interview), spec writing, dependency-ordered task breakdown, ADR/glossary capture	Before writing code for any new feature, vague requirement, or architecture decision
tdd	Red-green-refactor loop, incremental one-step-at-a-time delivery, regression tests for bugs	Implementing any logic or fixing any bug
debug	Root-cause debugging: reproduce → hypotheses → minimal verification → fix → regression	Tests fail, builds break, behavior doesn't match expectations
verify	Iron law (no completion claims without fresh evidence) + six-dimension pre-commit self-review	Before claiming "done", committing, or merging. Never skippable
git	Small logical commits, conventional messages, solo-friendly branch strategy, merge decisions	Any commit, branch, merge, or conflict resolution
ship	Pre-launch checklist, rollback plan, post-launch verification, minimal production observability	Preparing to deploy to production, or when production needs logging/monitoring
api	REST design, unified error format, pagination/versioning, deep module boundaries, contract-first	Designing APIs, module boundaries, or frontend-backend contracts
secure	OWASP distilled for a small company: secrets, injection, auth/session, IDOR, dependencies	Handling user input, auth, storage, secrets, or external integrations
handoff	Compresses a session into a handoff document (goal, state, next steps, suggested skills), redacted and saved to temp dir	Switching sessions/agents, pausing work, or handing off to a sub-agent
delegate	Dispatches isolated sub-agents by spawning pi -p processes (task file + timeout + output file)	2+ independent tasks that can run in parallel, or isolated-context exploration

Each skill is a standalone directory (skills/<name>/SKILL.md) with optional assets/ templates and scripts/ helpers — see the Agent Skills layout.

Installation

Skills are installed per-agent by copying them into a global skill directory. Two options:

Option 1 — pi (~/.agents/skills/ is discovered automatically):

git clone https://github.com/wuxiangru915/Whip.git
cp -r Whip/skills/* ~/.agents/skills/

Option 2 — Claude Code / Codex (point the harness at the folder, or copy into ~/.claude/skills / ~/.codex/skills):

git clone https://github.com/wuxiangru915/Whip.git
cp -r Whip/skills/* ~/.claude/skills/     # or ~/.codex/skills/

Verify by starting a session and asking the agent to list its available skills — you should see the 10 Whip skills.

Compatibility

• pi — tested; ~/.agents/skills/ is auto-discovered
• Claude Code / Codex — follow the Agent Skills integration guide
• The delegate skill spawns sub-agents via pi -p; on other harnesses, replace the spawn command with your harness's subagent mechanism

Contributing

Skills are personal by nature, but improvements to the shared methodology are welcome — open an issue or PR.

Acknowledgements

Whip's methodology is distilled and deeply rewritten from three MIT-licensed projects:

• mattpocock/skills — grilling/domain-modeling/handoff ideas
• obra/superpowers — brainstorming, TDD, debugging, verification-before-completion discipline
• addyosmani/agent-skills — spec-driven development, security, shipping, observability coverage

License

MIT