v0.82.0

Fixed

• A new worktree starts from the remote, not a stale local checkout. Asking for a new worktree of a project now fetches and branches from the remote's default branch, says which commit it started from, and warns when it has to fall back to the local HEAD; the fetch never prompts for credentials. Before, a worktree could start days behind and the agent would work on old code without noticing. (#1173)

• Files Chimera writes keep their line endings. The agent's file writes went through a text mode that turned every newline into CRLF on Windows, so files it created disagreed with the repository and a translation hash passed on Windows and failed in CI; writes are now byte-exact, a new file gets LF, an existing CRLF file keeps CRLF, and translation hashes ignore line endings. (#1172)

• A missing verify tool no longer marks a turn as failed. When the check after a turn cannot run because its program is not installed in the checkout (say, vitest in a fresh worktree with no node_modules), the turn reads as not verified, with the reason and the fix (a setup command in .chimera/verify.toml), instead of failed; a check whose tests ran and failed still reads as failed. (#1170)

• The test suite never calls a paid provider or leaks a key. A text-to-speech test fell back to OpenRouter when a key was present, made a real paid call and wrote speech.wav into the repository; and gateways exporting a provider key left it in the environment for later tests. Both are isolated now. (#1169)

• A failed memory extraction is no longer silent. When the extraction behind a Code turn cannot start, the app now logs a warning naming the error type (never the conversation or a key), and the turn still answers. (#1168)

• The installed app keeps its keys whatever folder it is opened from. The backend read the .env of the folder the app was launched from and recorded that folder for the background service, so an app opened from a terminal elsewhere started with no provider key and every turn failed; the Start menu shortcut hid it. A packaged build now always runs its backend in the install folder. (#1167)

• A number tweak applies when you finish typing. The field applied on every key, so typing 34 into a tweak that runs from 28 to 56 sent 3 first, was refused and snapped back; it now applies once, on leaving the field or on Enter, clamps to the range, and Esc discards the draft. (#1166)

Changed

• The README no longer calls Chimera self-evolving. Seven pre-registered learning runs found no measurable gain (bench/learning_lift/RESULTS.md), so the headline, the package description, the CLI help, the app and the translated READMEs and docs now say what the evidence supports: a governed agent that publishes its own null results. The translated READMEs also no longer promise an A/B test that the learning loop does not run. (#1171)

Added

• Chimera asks before working where commands are off. A coding request in a folder whose commands are off now stops before any tool runs and asks the owner to allow them, continue without, or cancel; only the owner's screen can grant, a turn with nobody at the screen never waits, the choice is remembered per conversation, and an unverified answer says so at the top. Before, the agent could edit for an hour and only then say it could not check anything. (#1175)

• The bridge pages and filters the conversation list. Listing conversations through the Claude bridge returned all 1,465 at once (about 1 MB); it now takes limit, cursor, state, project and q, defaults to 50 per page on the bridge, and the screen's unparameterised list is unchanged. (#1174)