claude-hooks

Produce a complete, restore-ready state summary of the current session. Use this when the user wants to compact context, pause the session, or hand off work — especially with phrases like "compact the context", "save state", "wrap up", "let's continue another time", "we will continue", "context is running low", "/wrapup", or any formulation that asks to end/pause the current conversation so it can be resumed later.

Configures the claude-hooks LSP engine for the current project. Phase 0 audits cclsp.json against the project's actual language files and proposes additive LSP entries. Phase 1 proposes [compile_aware.commands] for .claude-hooks/lsp-engine.toml by detecting build tools (Cargo.toml, tsconfig.json, pyproject.toml, go.mod, Makefile). Use when the user wants Claude to set up or extend LSP coverage or wire pyright / gopls / rust-analyzer / clangd / cargo check / tsc --noEmit / mypy into the LSP engine.

Multi-agent council engine (v2: planner → researcher → critic → synthesizer, plus opt-in tool_executor + coder; CitationLinter verifies every path:line). Default verb is `ask <question>` (also implicit — `/consultants <question>` works). Subcommands — `ask` runs a fresh council on a question; `followup [sid] <question>` iterates on a prior consult (warm reuse of plan/research/critic); `list` shows past sessions; `show <sid>` re-reads a stored summary; `config [args...]` walks the role/model/effort/service-mode dialog. Use when a question benefits from independent specialist agents working in parallel — design audits, release-notes validation, complex bug triage, refactor risk analysis. For single-shot questions use /get-advice instead.

LLM-to-LLM advisor — runs a multi-turn conversation with a configured Ollama model to get a focused second opinion on a question, validation request, or technical review. Default verb is `ask <question>` (also implicit — `/get-advice <question>` works). Subcommands — `ask` runs / continues a session; `model [NAME [CTX]]` reports or sets the advisor model and context length; `effort [tier]` reports or sets the sessions-per-invocation budget (`low`/`medium`/`high`/`max`); `tools [spec]` reports or sets the tool list exposed to the advisor (CSV / `all` / `none`). For multi-agent cross-cutting questions use /consultants instead.

Search past Claude Code conversations using episodic-memory. Works from any host by querying the remote episodic-server API. Use when the user asks 'I did this before', 'how did I fix X', 'find the session where', 'search my history', or references past work.