kochetkov-ma

Optimizes text, prompts, and documentation for LLM token efficiency. Applies 41 research-backed rules across 6 categories: Claude behavior, token efficiency, structure, reference integrity, perception, and LLM comprehension. Use when optimizing prompts, reducing tokens, compressing verbose docs, or improving LLM instruction quality.

Converts design screenshots or text descriptions to HTML/CSS using GLM vision API (Z.ai). Trial version — HTML output, CREATE mode. Full version with React, Flutter, review/fix modes: brewcode plugin. Triggers: design to code, screenshot to html, mockup to code, d2c trial, convert design.

Orchestrates end-to-end testing workflow — sets up specialized agents, creates BDD scenarios, writes Playwright autotests, and runs quorum reviews. Modes: setup, create, update, review, rules, status. Triggers: e2e tests, end-to-end, BDD scenarios, write autotest, playwright e2e, create scenario, review e2e, e2e setup.

Checks prerequisites and analyzes project structure to generate adapted templates in .claude/tasks/templates/.

Creates and manages dynamic teams of domain-specific agents (5-20 agents) for a project. Analyzes codebase, proposes team, scaffolds with tracking framework. Modes: create, update, status, cleanup. Triggers: create team, agent team, domain agents, team of agents for project, update team, team status, cleanup team.

Generates AI images via OpenRouter (Gemini/GPT-5), Z.ai GLM-image, Google Imagen 4, or OpenAI DALL-E 3 with anti-AI-slop controls. Modes: generate, edit, config, update. Triggers: generate image, create image, AI image, og image, blog image, illustration.

Disables or enables individual plugin agents by renaming <name>.md and _<name>.md in the plugin cache. Persistent state: global (default) or project. Triggers: disable agent, enable agent, toggle agent, hide agent, agent-toggle.

Analyzes project to extract etalon classes, reference patterns, and architecture by layer. Generates convention documents in .claude/convention/ and organizes project rules into a code style guide. Triggers: extract conventions, find etalon classes, analyze project patterns, code style guide, architecture conventions, convention docs, reference implementation, coding standards.

Executes a brewcode task PLAN.md with infinite context and automatic session handoff (third pipeline step after spec and plan). Triggers: start brewcode task, execute PLAN.md, run task with handoff, infinite execution, brewcode start, execute brewcode plan, run the plan.

Scans all git-tracked files for leaked secrets and credentials. Use when - scanning for secrets, security audit, finding leaked credentials. Trigger keywords - secrets scan, find credentials, security scan, leaked keys, security audit.

Toggles terse-output mode for Claude Code - injects brief directives into main conversation (SessionStart) and sub-agent prompts (PreToolUse:Task) to cut preamble and filler. Profiles: light, medium, aggressive. Scope: global or project (project overrides). Triggers: think-short, be terse, think shorter, reduce tokens mode, level 1/2/3.

Runs evidence-based multi-agent debates with mandatory Discovery phase (codebase + web research). Three modes: Challenge picks best variant from options, Strategy produces independent proposals via deep analysis, Critic finds all weaknesses. Triggers: debate, challenge this, compare options, strategy analysis, critique, find weaknesses, second opinion, pros and cons, evaluate alternatives.