mrkhachaturov

User-invocable skill for when Claude is stuck mid-task. Claude narrates its own blocker to Codex, gets back a structured diagnosis + concrete next steps, applies them, verifies, and loops up to N rounds until the problem is resolved or the loop detects it's spinning. Read-only — Codex advises, Claude edits.

Explore ideas before implementation. Turn vague requests into fully formed designs and specs through collaborative dialogue. Use before any creative work — creating features, building components, adding functionality, or modifying behavior. Proactively suggest when the user jumps straight to code without a design.

Write comprehensive implementation plans from specs or requirements. Use when you have a spec or requirements for a multi-step task, before touching code. Plans are bite-sized, TDD-first, with exact file paths and complete code in every step.

Internal helper contract for calling the codex-companion runtime from the codex-rescue subagent. Not user-invocable.

Internal contract for presenting Codex output back to the user. Referenced by dual-review (structured findings) and rescue (task output). Not user-invocable.

Sequential Claude-Codex review loop for specs and plans with structured JSON findings. Embedded in brainstorming and writing-plans for automatic Codex review after self-review. Also user-invocable for manual ad-hoc review. Max 3 rounds default, early exit if clean.

Internal guidance for composing Codex / GPT-5.4 prompts used by rescue and dual-review. Covers XML block structure, grounding, output contracts, verification loops. Not user-invocable.

Delegate a substantial investigation, diagnosis, or implementation task to Codex via the codex-rescue subagent. Use when Claude is stuck, wants a second implementation pass, needs deeper root-cause investigation, or should offload long-running work. Supports background execution, model selection, and thread resume.

Look up Cursor (the AI code editor) documentation in the local Miyo-indexed folder labeled `cursor`. Use when answering questions about the Cursor editor, Agent and agent tools, Plan/Debug/Design modes, Tab, models and pricing, rules, MCP, hooks, skills, plugins, subagents, the CLI, cloud agents, configuration, integrations, enterprise/admin setup, the Cursor SDK, or account/billing.

Look up Claude Code documentation in the local Miyo-indexed folder labeled `claude-code`. Use when answering questions about Claude Code — features, configuration, hooks, permissions, settings, MCP, plugins, skills, sub-agents, Agent SDK, CLI flags, slash commands, env vars, cloud providers, or IDE integrations.

Look up OpenAI Codex documentation in the local Miyo-indexed folder labeled `codex`. Use when answering questions about the Codex CLI, Codex app, IDE extension, cloud threads, AGENTS.md, config.toml, MCP servers in Codex, skills, plugins, hooks, sandboxing, approvals, slash commands, the Codex SDK, app-server, or the Codex GitHub Action.

Look up OpenCode documentation in the local Miyo-indexed folder labeled `opencode`. Use when answering questions about the OpenCode terminal UI, the `opencode` CLI, agents and subagents, opencode.json / opencode.jsonc config, skills, custom tools, plugins, MCP servers, providers and models, permissions, keybinds, themes, the SDK, share links, IDE integrations, GitHub/GitLab actions, ACP, or the LSP integration.

Look up Pi coding-agent documentation in the local Miyo-indexed folder labeled `pi`. Use when answering questions about the Pi terminal coding harness, the `pi` CLI, extensions, skills, prompt templates, themes, pi packages, providers, models, settings, keybindings, sessions, compaction, the Pi SDK, RPC mode, JSON event stream, or TUI components.

Cross-agent shared memory for YOUR_NAME. Save facts about YOUR_NAME that any agent should know. Reading is automatic via hook — your job is to write new facts.

Apple Health database — aggregates data from all sources (Apple Watch, iPhone, WHOOP app, third-party apps, manual entries) via Health Auto Export app. PRIMARY use: steps, walking distance, stand hours, active calories, VO2max, workouts, body weight, blood oxygen spot checks, nutrition (dietary calories, protein, carbs, fat, fiber, sugar) — anything logged to Apple Health. SECONDARY (prefer whoop skill when available): HRV, resting heart rate, sleep duration. Do NOT use for recovery score, strain, or sleep stages — use the whoop skill for those.

WHOOP CLI — authoritative source for recovery, sleep, HRV, RHR, strain, SpO2, skin temperature, and respiratory rate. WHOOP is worn 24/7 including during sleep, making it the most accurate source for overnight metrics. Use for: recovery score, sleep score, sleep stages (deep/REM/light), HRV, resting heart rate, daily strain, sleep debt, health insights, charts. Do NOT use for step count, walking distance, VO2max — WHOOP API does not expose these; use hae-vault for those. Trigger on: "recovery", "how did I sleep", "sleep score", "sleep stages", "HRV", "resting heart rate", "strain", "health dashboard", "health trends", "show me a chart", "health insights", "whoop".

Ask any question about Claude Code — CLAUDE.md, memory, hooks, skills,

Audit any project for AI-readiness. Analyzes Claude Code structure and suggests improvements. Run /audit-project from within the project you want to audit.

HindClaw documentation for AI agents. Use this to learn about access control, Terraform provider, multi-bank recall, session context, and configuration for Hindsight memory infrastructure.

This skill should be used when the user asks about the Model Context Protocol — "what does the MCP spec say about X", "look up SEP-991", "how does Streamable HTTP work", "MCP Apps SDK lifecycle", "JSON-RPC method shape", "capability negotiation", "OAuth + RFC 8707", "session management", "completion/complete", "resource templates", "sampling", or any wording in the spec or extensions. Provides semantic search over the locally mirrored MCP specification (all versions), SEPs, Apps SDK / Auth / Tasks extensions, and Anthropic MCP blog posts. Use BEFORE WebFetch'ing modelcontextprotocol.io — the local mirror is fresher and rate-limit-free.

This skill should be used when the user asks about the Model Context Protocol — "what does the MCP spec say about X", "look up SEP-991", "how does Streamable HTTP work", "MCP Apps SDK lifecycle", "JSON-RPC method shape", "capability negotiation", "OAuth + RFC 8707", "session management", "completion/complete", "resource templates", "sampling", or any wording in the spec or extensions. Provides semantic search over the locally mirrored MCP specification (all versions), SEPs, Apps SDK / Auth / Tasks extensions, and Anthropic MCP blog posts. Use BEFORE WebFetch'ing modelcontextprotocol.io — the local mirror is fresher and rate-limit-free.

Apple Health database — aggregates data from all sources (Apple Watch, iPhone, WHOOP app, third-party apps, manual entries) via Health Auto Export app. PRIMARY use: steps, walking distance, stand hours, active calories, VO2max, workouts, body weight, blood oxygen spot checks, nutrition (dietary calories, protein, carbs, fat, fiber, sugar) — anything logged to Apple Health. SECONDARY (prefer whoop skill when available): HRV, resting heart rate, sleep duration. Do NOT use for recovery score, strain, or sleep stages — use the whoop skill for those.

WHOOP CLI — authoritative source for recovery, sleep, HRV, RHR, strain, SpO2, skin temperature, and respiratory rate. WHOOP is worn 24/7 including during sleep, making it the most accurate source for overnight metrics. Use for: recovery score, sleep score, sleep stages (deep/REM/light), HRV, resting heart rate, daily strain, sleep debt, health insights, charts. Do NOT use for step count, walking distance, VO2max — WHOOP API does not expose these; use hae-vault for those. Trigger on: "recovery", "how did I sleep", "sleep score", "sleep stages", "HRV", "resting heart rate", "strain", "health dashboard", "health trends", "show me a chart", "health insights", "whoop".