mellions-deep-researchlisted
Install: claude install-skill LetA-Tech/mellions-coxen
<!-- Mellions Engineer | LetA Tech Ltd. | leta@letatech.ca -->
# Establishing what is true
Research is input, never the answer. The goal is not the most evidence but the
smallest sufficiently complete picture of reality a decision can stand on; how
far to go follows the stakes and the contradictions.
## Start from the question
Name what has to be true for the intended action to be right. That question
chooses the sources; a fixed list does not. "Is the defect still there at HEAD"
wants the code and a reproduction; "why does production do this" wants runtime
evidence.
## What exists
Kinds, not a checklist — reach for what the question needs:
- **the artifact, run and observed** — source, configuration, schema and
infrastructure at a named commit; the build, the tests, a reproduction, a
mutation; logs, metrics, traces, persisted state, captured requests, and
production where it is reachable within what is delegated;
- **history and the tracker** — `git log` and `blame`, prior fixes to the
path, the issue and its comments, related pull requests: accounts, with the
code as the referee;
- **governing documents** — intent, architecture, contracts, specifications,
and the repository's own statement of which has authority;
- **the sibling repository** when cause or consequence crosses a boundary,
pinned at its own commit;
- **other sessions' records** — assignments, handoffs, shift streams: what
they concluded and what they actually ran, which differ;
- **outsi