A small set of high-signal skills for software product work.
WyStack Agent Kit adds judgment where a capable coding agent benefits from a specialized frame. It does not impose a lifecycle, task tracker, workspace format, role hierarchy, or orchestration system.
| Area | Skills |
|---|---|
| Product definition | brainstorm, prd, story, glossary |
| Technical decisions | spec, adr, breakdown, estimate |
| Engineering quality | cross-review, verify, retro, improve-codebase |
| Product experience | frontend-design, ux-writing, copywriting, discoverability |
Each skill owns a distinct transformation:
brainstormturns an unclear idea into a reasoned direction.prddefines product outcomes, scope, and success.storyexpresses a user need as testable behavior.glossarygives a domain term one canonical meaning.specturns requirements into an implementable technical design.adrrecords a consequential architectural decision and its tradeoffs.breakdowncreates independently valuable work slices.estimatesizes work from scope, risk, and uncertainty.cross-reviewruns independent, evidence-grounded review angles and returns an explicit gate result.verifytests runtime claims with proportionate evidence.retroturns completed work into actionable learning.improve-codebasefinds high-leverage structural improvements.frontend-designexplores distinct static directions, then implements the chosen interface.ux-writingclarifies interface actions, states, and recovery.copywritingwrites persuasive product and marketing copy.discoverabilityimproves how a product is found by people and agents.
Use one skill or compose several naturally. The agent should follow the user's instructions and the project's existing conventions; a skill's built-in format is only a fallback.
cross-review is the one host-specific skill: it uses the Workflow runtime shared with pr-gate. On hosts without that runtime, use the host's native review.
- One skill earns its place by owning a distinct judgment or artifact.
- Descriptions are short routing signals, not compressed manuals.
- Skill bodies add only constraints and knowledge that materially improve the result.
- References load only when their branch is relevant.
- Completion criteria describe the outcome without scripting every move.
- Project and user templates override bundled defaults.
The writing approach is influenced by Matt Pocock's writing-for-agents: minimize always-loaded context, expose real decision branches, and remove instructions a capable model already follows.
/plugin marketplace add github.com/youhaowei/wystack-agent-kit
/plugin install wystack-agent-kit@wystack-agent-kitReload the session after installation.
codex plugin marketplace add github.com/youhaowei/wystack-agent-kit \
--manifest .agents/plugins/marketplace.json
codex plugin install wystack-agent-kitReload Codex after installation.
git clone https://github.com/youhaowei/wystack-agent-kit.gitPoint the host at plugins/wystack-agent-kit in the checkout.
grok plugin install github.com/youhaowei/wystack-agent-kit#plugins/wystack-agent-kit --trust
grok plugin enable wystack-agent-kitgit clone https://github.com/youhaowei/wystack-agent-kit.git
cd wystack-agent-kit
pi install .See CONTRIBUTING.md for local deployment and validation. See UPSTREAM.md for source lineage and the concept-port policy.
MIT