Targeted local agent/helper checks
Actually executed on the development Linux host, not the target VM:
/home/Kevin/.cargo/bin/cargo test -p rmm-agent -p rmm-agent-linux -p rmm-privileged-helper
First-increment result: 24 unit tests passed (4 agent, 12 Linux provider, 6 helper verifier/ledger/clock, 2 helper peer-ACL/policy). Tests live next to their private implementation APIs. They cover SQLite reopen, exact duplicate/conflicting offers, request reservation ambiguity, state/event rollback, cancellation/ expiry evidence, duplicate grants and attempts, unknown resource retention, every permit/identity binding, forged signatures, payload tampering, clock window endpoints, unsafe paths/corruption, peer UID mismatch, root execution denial in the network agent, actual harmless script bytes/output/limits and provider process-group cleanup. Unsafe service/package/reboot effects use existing typed mocks; native APT mutation is specifically tested as denied.
Also actually passed for these packages: cargo clippy --all-targets -- -D warnings,
cargo fmt -- --check, and debug cargo build -p rmm-agent -p rmm-privileged-helper
(using the explicit /home/Kevin/.cargo/bin/cargo path). Initial Clippy errors in
two provider test slice clones, helper accept-loop style and agent nested-if
style were corrected, not suppressed.
This is not a real browser-approved WSS job, VM service, package or reboot
acceptance. The complete bootstrap and known hard blockers are in
deploy/linux/README.md. Do not run native host mutation to replace VM evidence.
Second increment — actually executed locally
The targeted Rust command above now passes 34 tests (10 agent, 15 Linux, 7 helper-library, 2 helper-binary). Added tests cover native observation wire mapping/size/unsupported metadata, fresh live inventory time, typed log request allowlist/window/limits including inward fractional-time rounding, future-schedule fairness, write-once signing-key pins, actual loopback mTLS key bootstrap with DNS/SNI-preserving connect override, wrong-server-name rejection without creating trust, and durable original boot/ awaiting-reboot evidence with retained resource locking.
Also actually passed:
python3 tests/agent_jobs/test_install.py
python3 -m py_compile deploy/linux/install.py tests/agent_jobs/test_install.py
Two create-only staging/guard tests passed; the installer was not run against
the host root or VM. An initial Python annotation incompatibility on the local
interpreter was corrected using postponed annotations; tests then passed.
Targeted Clippy with -D warnings, debug builds and Rust formatting checks also
passed. No native service/package/reboot effects, VM access or server restart.