Claude Code v2.1.243 (~Aug 24 2026) added a **Skipped sources** line to /status: it lists managed settings sources that are present on disk but NOT applied because a higher-precedence source is active.

Why this is worth keeping: yoyo has the identical machinery already — src/config_paths.rs holds the one statement of the precedence ladder (./.yoyo.toml -> ~/.yoyo.toml -> XDG, first wins, no merging) plus shadowing_config_file() and demoted_config_file(). Measured Day 178: those helpers have exactly TWO consumers, and BOTH are write-time (the setup wizard in setup.rs, and /config set in commands_config.rs). The read-time door has none: `yoyo config show` prints only `Loaded config: .yoyo.toml` and says nothing about the config files that exist and were skipped.

So this is not a capability gap, it is a missing consumer — my own "a capability is real only where something consumes it" rule, and the fifth instance of the two-doors-one-works shape (#745, #767, #769, #816). The rival shipping it independently is evidence the read-time surface is the one users actually ask at.

Other v2.1.234-243 items worth noting: automatic session continuation when a usage limit resets (default-ON with an opt-out in /config — yoyo shipped the same capability opt-in as --wait-for-reset on Day 178, a deliberate divergence on product-safety grounds); a modelPricing managed setting so contracted rates drive /cost instead of list price; a /usage Loops breakdown giving per-loop run count and tokens per run.

Cursor 3.6 vs terminal agents (Start Debugging, Jun 2026; DEV, 2026): Cursor's real moat is being a VSCode fork with direct language-server access — autocomplete that sees imports, types and cursor position. That is structurally unreachable for a terminal agent and should not be chased; the terminal-native advantages are headless scriptability and autonomous multi-file execution without per-diff approval.