jr-claude-toolkit

Fast, focused review of uncommitted changes, guided reading tours of PRs and of implementation plans, plus a security-focused subagent.

4 skills · 2 agents

Advises on computation form with cheap grounding, exact and typed design, bounded residual composition, and experimental mixed planning; preserves pending work and model-router independence.

5 skills

Re-narrates a feature branch before merge: rebuilds messy exploratory history into a semantic commit sequence, with a byte-identical final tree and per-commit build gates. Forward mode (narrate-wip) turns the dirty working tree into semantic commits as you go, without rewriting history.

3 skills · 1 agent

Recommends the cheapest model, effort level, and execution shape that will complete a task reliably — routing atomic tasks with choose-model and decomposing multi-model work into routed execution plans with plan-execution — from versioned per-provider routing policies (Anthropic and OpenAI) and a project-local calibration log.

4 skills

Right-sized architecture documentation: scaffolds a lean docs/architecture layout sized to the repo's real risk, drafts ADRs by interviewing the human for the why it cannot infer, and a read-only drift reviewer that flags diffs touching documented boundaries, contracts, or decisions.

2 skills · 1 agent

Role-shaped delegation to Codex models (gpt-5.6 luna/terra/sol) from inside Claude Code: a general delegate skill plus QA and UX subagents that package this session's context into a self-contained prompt, run the analysis on a cheaper model via `codex exec`, and return findings as verified claims.

1 skill · 2 agents

Pre-deploy checklist, changelog generation, and a guard hook against destructive shell commands.

1 skill · 1 command · hooks

Compiles a raw intent into a prompt anchored on canonical terms the model already holds — scoring each term for canonicity, frame precision, frame fit and shared ground, gating the result on what a wrong expansion would cost, and naming everything the compression removed. Ships the mirror skill that reads a compressed prompt back and itemizes the entailments it imports, plus a project lexicon that records how each anchor actually behaved.

3 skills

Harvests durable knowledge from a session (or from topics you name) into notes: triage of concepts, systems and mechanisms worth keeping, routed to review only (the default), budgeted research in a read-only agent when a reason is named, or recording what the user stated; claim-by-claim source review in a clean-context agent; evidence levels (stated, verified, corroborated) on every note; --deepen to corroborate a note later; and an approval gate before anything is written. Plain markdown store by default; Basic Memory or any MCP-backed store by config; sources are whatever the project declares or has connected.

1 skill · 2 agents

Lifecycle review of code comments: verifies them against the code they describe, deletes noise, keeps load-bearing constraints. Diff-scoped by default, path-scoped on demand.

1 skill · 1 agent

Upstream jev-code integration pinned by this marketplace: Jev typed judgment tools over MCP plus the /jev skill, with Claude Code and Codex support.

1 skill · 1 MCP server

Upstream lavish-axi skill pinned by this marketplace: turns plans, comparisons, diagrams and reports into HTML pages opened in a local browser, where you annotate elements and send the feedback back to the agent. code-review's pr-tour and plan-tour use the same CLI for their --html mode.

2 skills

Runs incremental Stryker.NET mutation testing only on changed C# behavior with verified existing unit tests, using temporary tooling and configuration.

1 skill

Rewrites the last response in one or more named writing styles (STE, plain language, BLUF, ELI5, dev docs, visual) for side-by-side comparison, with optional readability scores and per-project style preferences — plus a native Visual output style that makes Claude converse in diagrams by default.

1 skill

Plans and applies a repeatable GitHub repository standard: protected linear history, PR-only integration, automatic Copilot review, and optional quality gates.

1 skill

Session organization helpers: distills the current session into a short, searchable resume handle and hands back the /rename command to apply it.

1 skill

Official TypeSafe upstream skill pinned by this marketplace: design typed System One/Jev judgments, question schemas, workflows, and API usage.

1 skill

Verifies UI behaviour in a browser through the Playwright MCP and leaves evidence a reviewer can trust: agreed test cases, safe WebM video with a chapter per case, DOM reads before and after each action, screenshots, a PASS / FAIL / not covered README and a timeline. Records natively with the MCP's video tools or through a script.

1 skill