Carinya Parc Agent Skills for Digital Product Delivery

Opinionated skills that guide an AI agent through the full product delivery loop — from strategy and architecture to epics, implementation, review, and sprint-end refinement.

Each skill produces one clear artefact (a markdown file or code change). Skills chain together: the agent reads what you already wrote and knows what not to put in the wrong document.

Getting started

Ask your agent:

"Setup Carinya Parc skills locally from this repo:
https://github.com/carinyaparc/skills

Clone it and tell me:
- What skills I have and what they do?
- How can I use the new skills?

Install as a plugin

Cursor

git clone https://github.com/carinyaparc/skills.git ~/.cursor/plugins/local/carinyaparc-agent-skills

Claude Code

git clone https://github.com/carinyaparc/skills.git ~/.claude/plugins/carinyaparc-agent-skills
claude --plugin-dir ~/.claude/plugins/carinyaparc-agent-skills

Install skills from skills.sh

# All skills from this repo
npx skills@latest add carinyaparc/skills

# Or one skill at a time
npx skills@latest add carinyaparc/skills/code-review
npx skills@latest add carinyaparc/skills/sprint-planning
npx skills@latest add carinyaparc/skills/ux-design-review

Try your first commands

/product --stage pitch
/roadmap
/tasks --product
/design checkout-foundation --mode walking-skeleton
/tasks checkout-foundation
/implement CHK01-01
/code-review
/code-review-fix
/merge-request CHK01-01
/validate checkout-foundation

Not sure where to start? Use skills-index, or follow the typical flow below.

Skills overview

Stage	Key outcome(s)	Skills
Planning	What, why, and when?	product, roadmap, tasks
Architecture	How? Structure? Principles?	solution, adr
Discovery	Ready for Development	design, tasks, backlog-refine
Delivery	Definition of Done	implement, code-review, code-review-fix, ux-design-review, ux-design-fix, merge-request, ralph-loop
Release	Ready for Release	merge-request-review, merge-request-babysit, validate
Refine	What did we learn?	sprint-planning, sprint-retro, docs-review

Typical flow

        product → solution → roadmap → backlog
                        ↓
            design → tasks (+ ADR optional)
                        ↓
    implement → code-review → code-review-fix
   (+ ux-design-review → ux-design-fix for UI changes)
                        ↓
   merge-request (+ babysit) → merge-request-review
                        ↓
              validate (epic done?)
                        ↓
        sprint-retro, docs-review (ongoing)

Or run the whole delivery stage autonomously: /ralph-loop-setup {epic} then /ralph-loop start loops implement → code-review → code-review-fix → ux-design-review → commit per task, then epic review, validation, and the merge request — one step per iteration until the completion promise is genuinely true.

The loop is not limited to software. It ships three presets: engineering-delivery (the flow above), ad-hoc (repeat one prompt until done), and custom (your own steps). See preset authoring for a worked non-engineering example.

The ralph-loop skills are driven by this plugin's own stop hooks (hooks/), shipped for both Cursor and Claude Code. Loop state lives in .claude/loop/ or .cursor/loop/. If you also have a standalone Ralph loop plugin installed (e.g. ralph-loop-plugin or ralph-wiggum), disable it — two stop hooks firing per turn would conflict.

Where files live in your project

Default layout the skills expect (override paths in your prompt if your repo differs):

docs/
├── product/
│   ├── product.md
│   ├── roadmap.md
│   └── backlog.md
├── architecture/
│   ├── solution.md
│   └── decisions/
│       ├── register.md
│       └── ADR-NNNN-{title}.md
└── work/
    ├── checkout-foundation/     # epic
    │   ├── design.md
    │   └── tasks.md
    └── sprint-3/
        ├── plan.md
        └── retrospective.md

Epic slug {epic} — kebab-case from the epic title or short title, at most two words (Checkout Foundation → checkout-foundation). Epic IDs like CHK01 stay in the backlog table; resolve the slug from that row when invoking skills.

Full path and boundary rules: delivery conventions.

Skill catalogue

adr is the only skill left with modes; it takes the mode first (/adr plan CHK01). The rest take their argument directly (/tasks checkout-foundation, /sprint-planning 3).

Product Strategy

Skill	Modes	Description	Artefact
product	—	Pitch or full product.md (why, who, what); review via docs-review	docs/product/product.md
roadmap	—	Outcome-based phases with exit criteria; review via docs-review	docs/product/roadmap.md

Architecture

Skill	Modes	Description	Artefact
solution	—	Stub or full arc42-lite solution.md; review via docs-review	docs/architecture/solution.md
adr	plan, write, review	Proposals in register.md; accepted decisions as ADR-NNNN-{title}.md	register.md, ADR-NNNN.md

Discovery

Skill	Modes	Description	Artefact
design	—	docs/work/{epic}/design.md (walking-skeleton or TDD); review via docs-review	docs/work/{epic}/design.md
tasks	—	Decompose anything into delivery work: a product into epics, an epic or its design into stories and tasks with Gherkin AC, or a spec/RFC/PRD into both	docs/product/backlog.md, docs/work/{epic}/tasks.md
backlog-refine	—	Groom an existing backlog or judge sprint readiness: reprioritise, split, re-estimate, defer. Amends in place and reports a verdict	backlog.md, tasks.md

Delivery

Skill	Modes	Description	Artefact
implement	—	Implement a task against approved design and tasks	code
code-review	—	Review a branch, PR, or working diff against its acceptance criteria and declared scope. Read-only: writes a verdict, never source	code review
code-review-fix	—	Address findings from a code review without changing observable behaviour; runs the project's validation suite and commits	code
ux-design-review	—	Live-first UX review of implemented UI vs its design source (Figma via MCP, mockups, tokens): accessibility (WCAG 2.2 AA), states, responsiveness, fidelity, design-system conformity. Read-only: drives the browser once, writes a verdict and captures, never source	UX review
ux-design-fix	—	Change how existing UI looks or behaves — from a UX review verdict or a direct instruction. Fixes via tokens and library components, re-renders to verify, re-checks neighbours, commits	code
merge-request	—	Open an MR/PR on any provider (GitHub, GitLab, Bitbucket) with a template-aware description. Never modifies source	MR / PR
ralph-loop-setup	(interview)	Seed and configure a Ralph loop: choose a preset (engineering delivery, ad-hoc, custom), resolve the environment, set the promise and iteration budget; writes the loop files, never starts them	seeded loop
ralph-loop	start, status, cancel	Run an autonomous loop: one step per iteration, plugin hooks re-feed the prompt until the completion promise is genuinely true or a safety rail fires	committed epic + MR

Release

Skill	Modes	Description	Artefact
merge-request-review	—	Review an MR/PR as its reviewer and publish inline comments and an approve / request-changes verdict; handles re-review rounds	published review
merge-request-babysit	—	Drive an open MR/PR to merge-ready: watch CI, fix objective failures, triage review threads, sync conflicts. Never merges	merge-ready MR
validate	—	Epic completion vs tasks and roadmap gates	validation report

Refine

Skill	Modes	Description	Artefact
sprint-planning	—	Plan a sprint: goal, carry-over, capacity, committed scope, dependencies, DoD	docs/work/sprint-{id}/plan.md
sprint-retro	—	Review a finished sprint: commitment vs actual, themes with evidence, actions routed to owning skills	docs/work/sprint-{id}/retrospective.md
docs-review	—	Review any set of documents: writing and structure per document, boundaries and duplication between them, consistency and cohesion across the set. Read-only	doc review
skills-index	—	“Which skill should I use?” for open-ended questions	routing

License

Copyright (c) 2026 Carinya Parc. Released under the MIT License.