watchthelight labs

Supersearch — Example Invocations

/supersearch is global. Invoke it from any project. It is always exhaustive — there are no quick/deep modes.

Enter mode (no topic) — prints the welcome, then waits

/supersearch

Start exhaustive research immediately

/supersearch whether React Server Components are production-ready in 2026

/supersearch best architecture for a local-first Electron + React + SQLite app

/supersearch Claude Code vs Codex CLI vs Cursor for autonomous software engineering workflows

/supersearch verify the claim that KVS Player V6 is commonly used for piracy or adult tube sites

Mode commands (while Supersearch Mode is active)

exit                          leave Supersearch Mode
cancel                        stop the running workflow, summarize what's done
help                          compact explanation of what the mode does
status                        current phase, agents returned, source/insight counts, what remains
refine: focus on self-hosting and ignore managed cloud options
sources: primary sources only — no vendor blogs
angle: evaluate everything from a security and supply-chain risk angle

Typical flow

  1. /supersearch <topic> → Phase 0 intent + Phase 1 plan (12–16 aspects) shown to you.
  2. Workflow fans out: aspect agents → source audit → contradiction hunt → timeline → practitioner reality → devil's advocate → synthesis-prep → quality gate.
  3. Quality gate verdict (PASS / WARN / FAIL) reported; one auto re-research pass on FAIL.
  4. Final dossier written (and optionally saved as supersearch-<slug>.md).
  5. Mode stays active — go deeper, change angle:/sources:, or exit.

Notes

  • Mid-research, drop a refine: / sources: / angle: line to steer without restarting from scratch.
  • High-stakes topics (legal, medical, financial, security, safety) automatically raise source standards and flag uncertainty.
  • A bare non-command message while in mode is treated as the research topic (or a follow-up if a dossier already exists).