feat: establish the canonical authenticated service boundary (DG1-P1)

[novelKR]

Behavior

Deliver DG1-P1 (DG1-C01 and DG1-C02): a canonical, exclusive foreground service boundary and a small authenticated local client. Normal paths come from the OS account, so changing HOME, XDG, a project setting or a command argument cannot create a second normal authority. Explicit bootstrap remains separate from ordinary startup; missing, corrupt and unsupported journals fail closed.

devguardd serve validates and owns the existing journal, authenticates native UDS peer UID/PID and separate consumer/admin credentials, and exposes bounded status communication. The protocol has strict version/field decoding, 64 KiB frames, 32 session workers and absolute 250 ms frame deadlines. Private credential handoff is tested across actual subprocesses, including closure before a later exec. Native registration, principals, workload leases and execution remain closed until the required P2/P3 evidence exists.

The two implementation commits preserve the DG-0 contract/journal format and original 44-test baseline. New daemon/client crates extend the explicit workspace dependency checks; contract/core remain independent of CodeSpace and Codex. English operations/contracts/planning and reviewed Korean counterparts describe the implemented boundary and subsequent work separately.

Validation

• Rust 1.95.0, one Cargo job and one test thread: full validator passed formatting, Clippy with warnings denied, complete dependency-graph checks and 83 Rust tests on local macOS. Three ignored subprocess helpers are invoked and asserted by their parent tests.
• Documentation checks passed: eight checker tests, 16 reviewed language pairs, 46 work units, 23 logical PR groups, dependency DAG and local links. The approved design checksum and Apache-2.0 license are unchanged.
• python3 scripts/qualify.py dg1-authority --offline: 19 functional cases. python3 scripts/qualify.py dg1-auth --offline: 20 functional cases. These are scoped subsets of the regression tests, not additional OS qualification claims.
• Actual canonical foreground smoke: explicit first bootstrap, native daemon/client identity, FD authentication, closed readiness, duplicate authority refusal, two start/stop cycles, same journal/lock preserved, no remaining endpoint and zero workload attempts.
• Failure coverage includes FIFO/shared/linked locks, incompatible configuration, credential substitution/leakage, forged peer claims, strict added-field rejection, final response after peer close, slow input/output, partial-frame saturation recovery, stale endpoint reconciliation and bounded shutdown. Independent review found no remaining blocker after the unsafe-lock fix.
• Existing macOS/Ubuntu CI validation remains required; both scoped functional suites and their reports are added to that matrix. Hosted current-head checks and separate post-merge main checks are verified as delivery gates.

Compatibility, limitations and rollback

Configuration schema remains 1, but system_tasks now requires at least 48 (32 bounded sessions plus 16 service/control tasks). The former C01 value 16 is explicitly rejected. An operator must review capacity and reservations before updating it; there is no automatic migration, capacity enlargement or journal reinitialization. This is an accounting estimate, not a kernel task limit.

Authentication is not isolation from malicious same-UID programs. Frame deadlines are functional transport bounds, not the later end-to-end admission or foreground SLO qualification. Linux CI covers portable contracts and local transport fixtures, not DG-LINUX enforcement. No LaunchAgent, generic execution, self-use, resource-control qualification or CodeSpace runtime integration is delivered here.

Rollback: stop the foreground process and select an artifact compatible with the preserved configuration and schema-1 journal. Preserve credentials, journal and evidence. C01/C02 cannot have launched user workloads; later live-lease rollback must use actual reconciliation rather than this early empty-state assumption.

Local reports, diagnostic evidence and foreground receipts are preserved outside the disposable worktree. After normal exact-head merge and separate main CI success, remove only this task's clean worktree, merged local branch and regenerable build output; retain the remote branch, operational state, credentials and evidence.