[paleo]

Changes

• Add installation identities and restrict interrupted-session sealing to conclusively dead runs on their owning machine.
• Keep running logs local as *.running.md, rename them on completion, and accept retained running paths in code status.
• Add clone-local sync locking, sync-only commit consolidation, and bounded retries for lock contention and remote advancement.
• Stage only resolved conflict paths, abort unresolvable rebases, and clarify recovery commands.
• Run Git non-interactively with default SSH and HTTP transport timeouts while preserving configured SSH commands.
• Update architecture, setup guidance, release notes, and regression tests. Included guide changes favor simpler specifications and leave PR draft preferences to workspace instructions.

Known issues and accepted scope

• Stale-lock takeover remains defective. A contender can rename a replacement lock created by another live process, so two processes can both acquire ownership. The earlier owner's release can then delete the later owner's lock. The user explicitly accepted proceeding with this known finding and requested opening this PR. The code is unchanged after that acceptance.
• The pre-existing aligndev code launch-admission race is separate. It remains outside this ticket's scope, with a side ticket deferred until after merge.

Validation

• Earlier execution validation passed repository-wide typecheck, build, lint, 2,149 unit tests, and 113 integration tests. Disposable built-CLI checks also passed at that stage. These results predate the final review fixes.
• The review-fix summary records passing AlignFirst tests (426), AlignDev tests (887), typecheck, build, lint, and docmap checks.
• Subsequent targeted validation rebuilt AlignFirst and passed 51 tests across sync-lock, Git, and conflict handling. A deterministic two-process probe reproduced the stale-lock defect above. Validation then stopped; the remaining repository-wide checks and built-CLI completion/retained-path and transport probes were not run.
• No complete final rerun is claimed. No OpenClaw conversation-model scenarios ran, and no real consumer repository was manually mutated.
• PR preparation confirmed matching local/remote branch tips at d838def and that main is an ancestor. A fresh branch-wide git diff --check reported existing trailing whitespace in three OpenClaw workspace instruction files; those files were preserved unchanged.