Slice 7 Delegated Baseline Smoke¶
Status: delegated-baseline-smoke-verified. Scenario:
matched_safe_read. Run:26571739019. Proof pack:assay-runner-delegated-proof-pack-26571739019(artifact7264883391, retained until 2026-08-26). Assay commit:c3384c425673fe09b0368f82765c72dda86ac200.
Outcome¶
The delegated openai-agents-kernel-policy gate passed all three deterministic OpenAI Agents kernel+policy runs. The proof pack records three Runner archive tarballs, selected JSON, gate logs, and four pass lines: one acceptance pass for each run plus the three-run determinism pass.
The delegated baseline is therefore a positive_join, not a semantic-gap finding. It proves the positive join path for the existing OpenAI Agents fixture under real Runner capture.
Evidence¶
| Check | Result |
|---|---|
| Gate | openai-agents-kernel-policy |
| Workflow inputs | gates=openai-agents-kernel-policy, build_ebpf=true |
| Runner health | kernel_layer=complete, ringbuf_drops=0, cgroup_correlation=clean |
| SDK evidence | one tool_call_started and one tool_call_completed for tc_runner_policy_001, tool read_file |
| Policy evidence | allow decision for tc_runner_policy_001, tool read_file |
| Kernel evidence | two successful workdir-bounded openat read events |
| Correlation | clean, one binding for tc_runner_policy_001, zero ambiguities |
| Join result | strong tool_call_id join, no fallback |
| Scenario verdict | positive_join |
Implementation Note¶
The first delegated attempts exposed a cgroup-root issue under the self-hosted assay-bpf-runner systemd service: using a .service unit as the session root can become invalid for child process placement once systemd reports the service cgroup as threaded. The fix in this branch treats .service units like .scope units and ascends to the nearest non-leaf domain cgroup before creating Assay session cgroups.
Non-Claims¶
- This smoke does not publish delegated semantic-gap findings.
- This smoke does not dispatch delegated gap scenarios.
- This smoke does not promote evidence packs, semantic-gap verdicts, or join rows to product APIs.
- This smoke does not rank Runner, OTel, OpenInference, or the OpenAI Agents SDK.