Pre-computed context

Current branch: !git branch --show-current 2>/dev/null || echo "unknown"

Purpose

External research is mandatory before acting on external facts. Goal: maximum knowledge, maximum consensus, latest information from authoritative + official sources. Training data drifts; library APIs change; SEO content farms outrank authoritative sources; AI synthesis tools repackage the same secondary blogs as "multi-source." Cross-tool consensus + primary-source priority + recency verification drive accuracy.

Local counterpart: /explore (what IS in the repo). This skill covers what SHOULD BE, per current external sources. For a multi-topic or workflow-driven pass, use /research-deep, which layers tiered execution on top of this discipline. Philosophy: more tokens + more time = more accuracy + less rework — insufficient research is the #1 source of rework. Deploy a research team, not a single lookup, and give every invocation full depth regardless of task size.

Routing — dispatch by default

From the main conversation, this skill dispatches the discovery:researcher subagent. Research reads a lot; keeping that out of the orchestrator's context window is the point. The agent runs Phase 0 through the gate's mechanical criteria, writes the artifact set, and returns a file pointer plus a short summary — not the transcript. The parent resolves the pre-dispatch envelope first (topic, memory-slice path, memory root, budget, capability flags) and owns the post-dispatch boundary after: re-surfacing open_questions, dispatching the sibling verifier, applying project fit itself, and writing both results back into the index — verification: pending is a statement that the producer may not self-grade, not a permanent state, and an artifact left saying pending after the parent verified it understates what is known to every later reader.

Run inline instead when any of these holds — and inline runs the identical discipline; the escape hatch relaxes nothing below:

Tight turn-by-turn iteration — you will redirect the queries as they land. Dispatch is a pre-run choice; the steering loss is mid-run.
Cost — a dispatched run pays full depth every time, including for a one-line version lookup whose doc you can already name.
The invoking context is already a subagent — dispatch-by-default is scoped to the main-conversation boundary, so a subagent invoking this skill runs it inline. The outer dispatch already supplied the fresh context. Hoisting, not nesting.