[igorcosta]

Fresh installations leave background sync off. Turning it off cancels the timer and in-flight or queued sync immediately; explicit opt-in can start it again during the same process. --sync-settings false retains the existing boolean parser. Sync toggles do not mutate shared default configuration.

Continuation chooses the most recently active conversation with messages in the selected project, with an empty-session fallback only when all candidates are empty. /sessions hides empty rows; /resume already does so.

Foreground run_command and shell observations sent to the model are capped at 30,000 characters across stdout and stderr, with head/tail output and an artifact path in the marker. Raw output exceeding the limit is spooled to a private session log before the independent 1 MiB terminal captures truncate it. tools.maxCommandOutputChars and AUTOHAND_MAX_COMMAND_OUTPUT_CHARS override the model limit. Streamed chunks and nested script dispatch cannot bypass the cap. Ephemeral sessions retain their no-persistence boundary, and short command-result shapes remain compatible.

This draft covers F-0399, F-0400 and F-0402 and targets the pending cto/release-0.9.10-final-20261104 branch. Matt's sync documentation sentence and Igor's maintainer review remain pending. Nothing has been merged, released or published.

Validation:

• 469 focused unit tests passed. Native TypeScript, complete ESLint and build passed.
• 16 focused built-CLI Tuistory scenarios passed, including the three-launch BANANA-42 probe with local authenticated fixtures. The two new scenarios were rerun with a distinct final-answer marker and passed.
• Full bun run proof: unit phase 13,797 passed, 4 failed, 80 skipped (964 files passed, 2 failed, 6 skipped). Its chained build/Tuistory phase did not run after the unit exit; build and focused terminal checks were run separately.
• Aggregate failures: three hook-context assertions in AgentDependencyComposer.outcomes.test.ts and the help-line case in AgentUI.test.ts. The same filtered cases pass in isolation on both release base and draft (6 passed each). Their aggregate cause remains unresolved; this is not a full green proof.

F-0402 live acceptance remains incomplete:

• Both real providers executed seq 1 400000 once and returned a confirmed 400000. Each model tool result was 29,950 characters and each full raw archive held 2,688,895 bytes / 400,000 lines. The 20k intact-log test passed.
• gpt-5.6-terra: 10.12 seconds, 40,871 actual prompt tokens, one confirmed reply.
• Fantail: 24.19 seconds, 71,720 actual prompt tokens, a provisional draft followed by the confirmed reply.
• The approximately 30k total-token target and Fantail single-reply behavior are not met. Additional context/completion-review work is still required. Do not mark this draft ready to merge.

The live probes used build 432734e. Subsequent commits change only terminal tests; production source remains identical. The process name remains autohand.

[igorcosta]

Incorporated into #750 and merged into main at e59017ba7ca7. The audited source commit 06d7eaee7c221f527d5d7a70a8ece8c7596ffac0 is represented in that validated integration, including the necessary integration corrections. Closing this source PR as incorporated; its branch is retained.