pablolok/claude-skills

pablolok's skills for Claude Code. The repository is a Claude Code plugin marketplace, pablolok-skills: every
skill is a plugin of its own, so a person installs any of them with Claude Code alone — no clone, no Python. A shared
or cloud project keeps managed copies instead (see below).
Skills
| Skill | Description |
| :--- | :--- |
| **Pre-Implementation Review** | Before writing code, decide the right design — responsibilities, the pattern and owner of each, the seams; reuse and low duplication follow from it. |
| **Skill Retrospective** | When a piece of work closes, rewrites the skills it used in place with what it taught. The close steps of backlog and jira-workflow invoke it; one hook (Node) logs the skills used. |
| **Backlog** | The project's work tracked in BACKLOG.md, one doc folder per substantial piece of work archived on close, with the doc gates and a two-way GitHub Issues mirror (Node scripts inside the skill; project values in .claude/backlog.json). |
| **Review Backlog** | The periodic review of the backlog: run the gates, clean up what has rotted, close with ONE proposal of what to do next. Pairs with Backlog. |
| **Clean-Code Standards** | The language-agnostic quality bar: single responsibility, injected collaborators, Tell-Don't-Ask, no god objects, no static business logic, strong typing, centralized literals, a reuse audit, fail-fast, no warning suppression. |
| **C# / OOP Standards** | The C#/.NET layer on top of Clean-Code Standards: idioms, tooling and the coverage bar; the persistence stack is opt-in per project. |
| **TypeScript / React Standards** | The TypeScript, React and Node/Deno layer on top of Clean-Code Standards: idioms, tooling and quality gates. |
| **Python Standards** | The Python layer on top of Clean-Code Standards: idioms for injection, typing, errors, paths, encodings and subprocesses, with the project's own runner, minimum version and checkers. |
| **Test Design Standards** | What a good test is made of, in any language: a bug fixed from a reproducing test, behaviour through the public seam, deterministic and isolated tests, the right doubles and level, flaky tests fixed or quarantined. |
| **Verification Gates** | Before handing work back: the required gates (tests, builds, compilers, linters, static analysis) are green on real evidence and nothing was silently removed. |
| **Diff Self-Review** | Before a commit or a "done", and when accepting another agent's work: read the real diff and trace every hunk to the ask, so no behaviour changes that nobody asked for. |
| **Jira Workflow** | Work tracked in Jira: the tracker updated first, status moved when the fact happens, points, hours and continuous worklogs on every ticket, dependencies as Blocks links, the total review and the Atlassian connector's pitfalls; project values come from the project's playbook. |
| **Writing Architecture Docs** | The design doc of a piece of work and the stable "how it is now" doc of an area: three questions, the skeleton, nine rules, one diagram per question. |
| **Mermaid Diagrams** | Mermaid diagrams that really render in the VS Code preview and on GitHub: when to pre-render to SVG, a validator and a renderer behind one stable project command (node scripts/mermaid.mjs check / render), the dead ends. |
| **Savvy Flow** | An orchestrated delivery flow (/savvy-flow:savvy-flow ): the session model plans, owns the design and reviews adversarially; five tiered worker agents (savvy-flow:savvy-light … savvy-fable) implement. Imported from johnnyvizz/claude-kit (MIT). |