You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Two opt-in AI usage sources, both off by default and never enabled by --yes alone. usage.claudeLimits shows your plan's real Claude limits: it
reads the Claude Code login token from the login keychain and asks Anthropic's usage endpoint. This is unofficial: Anthropic's terms do not allow
using that token in other tools, so the account could be restricted. usage.opencode shows OpenCode Go usage: it reads the console session cookie
from a Firefox-family browser profile inside your home folder (a private temporary copy, removed afterwards) and asks the OpenCode console, which
is not a published API. Requests go over https to api.anthropic.com and opencode.ai only, with no cross-host redirects; secrets are never
logged or cached. The installer has a step for each, with a panel that says what is read and sent and needs y to accept
(--claude-limits, --opencode-workspace, --opencode-profiles for scripts). See docs/CONFIG.md.
Fixed
AI usage no longer shows a percentage it cannot know. Claude Code's local logs have token counts but not your plan's limits, so a percentage was
only "of your own peak" and read 100 % whenever you were at it. Without usage.fiveHourTokens / usage.sevenDayTokens in config.json the atom
and card now show the tokens used (the running 5-hour session and the last 7 days); with them, a real percentage, as before.
The three-finger swipe could end up reversed until riskell restarted: its direction was guessed from the first Space change after a swipe, and a
Space change from another cause close to a swipe taught it the wrong way. The direction now follows System Settings > Trackpad > Natural scrolling,
read each time, and a Space that moves against a swipe is only noted in the log.