bobtat-plugins

Teaches Claude to write, review, and audit tests worth having — behavior-focused design, disciplined test doubles, browser and component testing, a plan-gated pipeline that turns a ticket or PR into tests of the described behavior, a check on whether pushed code still matches the description that specified it, and a repo-scale map of where test protection is weakest

7 skills · 4 commands · 5 agents

Interactive, human-in-the-loop QA walkthroughs of finished functionality — drives a browser, an API, and a CLI against a ticket's acceptance criteria and its merged PR, with an adversarial critic pass on both the behavior spec and the step plan, drafter/critic revision rounds, and a report with embedded screenshots. Not automated test-code authoring; that's what the testing plugin is for

5 skills · 1 command · 7 agents · 1 MCP server · hooks

Teaches Claude to build the evidence record a performance review needs before the evidence is gone — a session hook that distils each transcript into a redacted digest of your own prompts before Claude Code's 30-day purge deletes it, keeping the decisions and discarding the tool output where the secrets live; a sweep that reconciles git, PR, and review history against the journal and interviews the gaps; and authoring that turns dated entries into a review, a promotion case, or resume bullets without inventing a number

4 skills · 5 commands · 1 agent · hooks

Teaches Claude to model and check jewelry in Blender with the JewelCraft add-on over a Blender MCP connection — prong grip, seat clearance and stone overlaps measured rather than eyeballed, ring sizes read from the band's actual inner surface, metal weight with overlapping parts counted once, print/cast checks that confirm an STL comes out in millimetres, and metalwork built from editable paths fused with Geometry Nodes, split into a cast part and a hand-built wire head with a 1:1 bending template, with the add-on's scripting traps documented and worked around

11 skills

Teaches Claude to treat version control as a deliverable — atomic conventional commits split from a working tree, branch history made legible without interactive rebase, and pull requests that describe the change rather than the diff, with hooks that route version-control requests to it and block the irreversible mistakes

3 skills · 3 commands · 1 agent · hooks

Teaches Claude to edit its own prose — the catalog of tells that mark machine-written text, the clarity moves that rebuild a sentence around its real subject and verb, and the never-invent rules that keep an edit from becoming a rewrite, with a command that runs an edit pass over a draft and an agent that reads a finished document cold

2 skills · 2 commands · 1 agent

Adversarial review of a PR, plan or document by an isolated reviewer that never sees the author's context, with a triage pane deciding which findings reach the author

1 MCP server

Domain-driven design guidance for C# applications — aggregates, value objects, domain events, bounded contexts, and module structure

1 skill

GraphQL API design guidance and .NET implementation patterns for HotChocolate and GraphQL.NET — schema design, operations, Relay conventions, and performance

1 skill

MediatR guidance for in-process messaging in .NET — requests, notifications, pipeline behaviors, registration, and testing

1 skill

Teaches Claude to spot code smells (classic, architectural, and test smells) and apply disciplined, behavior-preserving refactoring

1 skill

Tracks what Claude Code costs — session, calendar-week and calendar-month totals with the month's pace and week-over-week change, a /costs pane with a spend calendar, the most expensive prompts and what filled the context, plus prompt-cache awareness: a cache warm/cold countdown, warnings before a prompt, model switch or large write re-caches the context, a per-category breakdown priced at list or your own (e.g. AWS) rates, per-project and subagent splits, budget alerts, CSV export and a /costs doctor self-check, the cache lifetime learned from Claude Code's own pricing, what a cold re-cache will cost before you resume, and an optional keep-alive that holds the cache warm while a session sits idle

A live /agents-tree pane showing the session's subagents as a tree — status, how long each has run, its tool calls and steps, the tool it is on now and the output tokens it used