core

Plugin core from marketplace dev-kit in risqaliyevds/claude-dev-kit. Commands read from the repo's manifest — 10 days ago.

Skills (20)

• ai-engineer

  Conventions for building LLM-powered features, agents, and AI pipelines. Use when writing code that calls model APIs (Anthropic, OpenAI, etc.), builds agents, RAG, or prompt pipelines.

• announcement

  Draft an internal team announcement (Telegram-style) in Uzbek, with optional Russian version. Use when asked to write or draft an announcement, e'lon, xabar, or internal message for the team.

• changelog

  Update CHANGELOG.md with recent changes in Keep a Changelog format. Use after completing a feature, fix, or refactor, when the user asks to update the changelog, or before committing significant work.

• commit

  Stage and commit current changes with a Conventional Commits message, updating the changelog first when appropriate.

• docs-sync

  Audit every doc against the current code and fix the drift - update stale claims, delete dead ones, archive finished plans. Use when docs mention removed features, wrong paths/commands, or old behavior.

• evals

  Create or run a small evaluation harness for an LLM feature, prompt, or pipeline. Use when building or changing prompts, RAG, agents, or any LLM-powered behavior, or when asked to eval, benchmark, or compare prompt or model versions.

• fastapi

  FastAPI and modern Python backend conventions. Use when writing or modifying FastAPI apps, routers, Pydantic models, SQLAlchemy code, or Python API endpoints.

• firstmate

  Act as my first mate - take directions, split them into independent tasks, run each as a background Claude session in its own worktree that implements and ships it, watch them, and report back so I only make the decisions.

• flutter

  Flutter and Dart conventions for mobile app development. Use when writing or modifying Flutter widgets, Dart code, pubspec.yaml, or building iOS/Android apps with Flutter.

• grill-me

  Interview the user relentlessly about a plan or design until you reach shared understanding, resolving each branch of the decision tree. Use when the user wants to stress-test a plan, get grilled on a design, or says "grill me". Asks in English, plus Uzbek when enabled.

• init-dev-kit

  Bring the current project fully up to the dev-kit standard - pull the latest kit, scaffold every standard file, triage and reorganize the whole tree (docs, scratch, secrets, artifacts) into the convention layout, and format. For onboarding an existing/messy repo, not just fresh ones.

• learn

  Turn this session's corrections into durable project memory and keep AGENTS.md lean - record recurring mistakes as one-line rules, and move instructions only some tasks need into project skills.

• new-project

  Bootstrap a new project with the standard setup - CHANGELOG.md, CLAUDE.md, .claude settings, and .gitignore.

• nlp

  NLP engineering conventions - text processing, datasets, multilingual (Uzbek/Russian/English) handling, and model evaluation. Use when working on NLP tasks like classification, NER, summarization, translation, embeddings, tokenization, or text datasets.

• overnight

  Run a long unattended agent loop (gnhf) toward a verifiable objective, with hard caps on iterations and tokens, in its own worktree - for overnight work such as raising test coverage, cutting load time, or fixing usability problems one by one. "review" summarizes a finished run.

• plan

  Deep planning and research before implementation - produce an implementation plan without writing code.

• release

  Cut a new release - move Unreleased changelog entries under a version heading, bump the version, commit, and tag.

• report-verify

  Verify declared figures in a report against the actual hyperlinks in the document - count links, group by domain, and cross-check claimed publication or content counts. Use when checking media reports, activity reports, or any document whose claimed numbers should match its links.

• senior-engineer

  Senior engineering standards for writing and changing code. Use when implementing features, refactoring, fixing bugs, or reviewing code in any language.

• ship

  Take finished work all the way to a verified PR (or a push to main where the repo says so) - intent, rebase, fresh-context review with auto-fixes, tests plus end-to-end evidence, docs and changelog, push, CI babysitting, risk level. Use only when the user, or the prompt that started this session, as

Subagents (2)

• code-reviewer

  Adversarial senior code reviewer working in a fresh context. Use after writing or modifying significant code, and as the review gate of /core:ship, to catch bugs, security holes, intent drift and unneeded code before a change ships. Returns findings tagged auto-fix or ask-user plus a risk level.

• researcher

  Deep research specialist. Use when a task needs heavy investigation first - exploring unfamiliar parts of the codebase, comparing libraries, or digesting long documentation - and only a concise summary should come back to the main conversation.

Hooks (2)

• SessionStart

  sh "${CLAUDE_PLUGIN_ROOT}/hooks/scaffold.sh"

• PostToolUse

  jq -r '.tool_input.file_path // empty' | xargs -r npx --no-install prettier --write --ignore-unknown 2>/dev/null || true

Similar plugins