Some checks failed
Build and Test / Build Server (Linux) (pull_request) Failing after 7m12s
Build and Test / Build Agent (Windows) (pull_request) Successful in 14m56s
Build and Test / Security Audit (pull_request) Successful in 7m57s
Build and Test / Build Summary (pull_request) Has been skipped
Address the SPEC-018 Phase 1 code review (reports/2026-06-03-spec018-review.md): - Bug 2 (config.rs): stop agent_id churn on every restart. The embedded-config path always wins in Config::load, so the saved agent_id was never read back. Add Config::persisted_agent_id() and reuse a prior id from the TOML; only mint a new UUID when none exists. - Bug 1 (main.rs): remove the non-functional in-process fallback in run_permanent_agent_managed. A managed agent's cak_ store is SYSTEM-only ACL'd, so a non-elevated in-process run cannot authenticate (load_cak permission-denied, or enroll C1 read-back failure). Return an actionable "install elevated" error instead of pretending to provide an agent; update the misleading comments. - Issue 6 (startup.rs): replace the fragile transmute::<HANDLE, HKEY> with the windows crate's typed HKEY out-param; add SAFETY comments. cargo check -p guruconnect --target x86_64-pc-windows-msvc passes clean. Deferred lower-severity items tracked in #8. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>