upstream-researchlisted
Install: claude install-skill flungo/claude-plugins
# Upstream Research
How Fabrizio wants facts about **third-party and upstream components** established: by going to the authoritative source, not by trusting recall.
Training data is frozen and often wrong on specifics; blog posts and web-search summaries are worse; even a project's *generated* docs can lie.
The cost of a wrong upstream fact is high — it gets baked into config or code and surfaces as a runtime failure much later.
Apply this whenever you're confirming how an external component behaves — a config option, an API or wire detail, an env var name, a version-specific change, a bug — whether or not a named command was invoked.
These are personal defaults that **complement repo/context rules, never supersede them**.
## The two reference files
- **`references/finding-sources.md`** — what counts as authoritative and how to read it (clone the source, fetch raw files, pin to a ref, navigate from the docs root), and how to record provenance so the next reader can re-check.
- **`references/what-to-distrust.md`** — the sources that mislead (training data, third-party posts and web-search summaries, generated docs, and archived or renamed predecessor repos), with the tell for each.