Claude Code cli.js Environment Variables: Complete Table

This is a revision of my earlier pass over Claude Code's bundled cli.js. The first version grouped variables into inventories, but the obvious missing piece was a full row-by-row table.

This version includes every environment variable I extracted from the bundle in one complete table, with a family label, a short explanation of likely purpose, and a note about whether the explanation comes from a directly visible startup path or from bundled-module inference.

For version 2.1.83, the extraction yields 477 unique environment variable names.

Method

I extracted all dot-form and bracket-form environment accesses from the bundled file and then deduplicated them:

rg -o "process\.env\.[A-Za-z0-9_]+" cli.js | sort -u
rg -o "process\.env\[['"][A-Za-z0-9_]+['"]\]" cli.js | sort -u

I also looked for assignments like process.env.FOO = ... so I could distinguish variables that the bundle writes or normalizes from variables it only reads.

How To Read The Table

• Likely purpose / usage is exact for the few top-level startup variables and intentionally conservative for the rest.
• Basis tells you whether the row is from direct bootstrap code or from a bundled-module naming inference.
• If you are debugging startup specifically, start with the Entrypoint family.
• If you are debugging auth, routing, or observability, the Anthropic, Cloud, MCP, and Telemetry rows are usually the high-signal ones.

Family Summary

Complete Environment Variable Table

Takeaway

The important distinction is still the same: the visible cli.js bootstrap surface is small, but the bundled reference surface is large because the artifact contains first-party product code plus cloud auth helpers, MCP, telemetry, runtime detection, platform probes, and vendored libraries.

Putting every variable into one table does not make the bundle a public contract. It does make it much easier to reason about which names are likely worth trying when you need to debug a concrete behavior.