smart-mcps

Executes an approved research_plan.md — grouping its questions across subagents to keep context tight — against external-knowledge sources (perplexity/notebooklm), writes a complete research_answers.md, then a concrete implementation_plan.md ready to hand to an implementer. Runs in a clean context, separate from /plan-to-plan, for full focus.

Chat with a NotebookLM notebook by topic. Ask questions and get grounded answers. Configure your notebook map in this file after install.

Full NotebookLM management — query, create, add sources, manage notes, and generate audio/video artifacts via the nlm CLI.

Think-from-base grilling that turns a vague idea into a requirements document with stable R-IDs, ready for /orchestrator-plan. Use when the user wants to think through what to build before any plan exists — purpose, constraints, success criteria, scope boundaries, failure modes, non-goals.

Produce an orchestrator-ready plan with an embedded task map — grounded in codegraph, grilled with the user, verified by a subagent — so `smart-mcps-orchestrate group` runs deterministically with the mapper LLM skipped. Input is a brainstorm requirements doc (from /orchestrator-brainstorm) or a direct feature description.

Launch, watch, and resolve an orchestrator run as the run-driver session — preflight the repo, start `smart-mcps-orchestrate run` detached with HITL on, triage every escalation yourself (fix env/config/plan and `retry`, `answer` from the docs, ask the human only for product decisions), and write up the outcome. Use when the user wants to run, resume, drive, or babysit an orchestrator run.

Create or maintain STRATEGY.md - the repo's target problem, approach, who it serves, key metrics, and tracks of work. Use when starting a new initiative, updating direction, or when prompts like "write our strategy", "what are we working on", or "set up the strategy doc" come up. Also relevant when /orchestrator-brainstorm or /orchestrator-plan need upstream grounding and no strategy doc exists yet.

Web-grounded search and reasoning using the Perplexity API. Use when you need current information from the web, or need AI-synthesized answers grounded in live sources.

Research planning — pins down the final objective, explores the codebase (codegraph-first) for what's already known, decomposes the remaining gaps into tagged external-knowledge questions (perplexity/notebooklm), and writes an approved research_plan.md (objective + resolved context + questions) ready for /apply-research-plan to execute.