Update sandbox-runtime to 0.0.79 and bind retained smoke evidence - #855

Scope

Update sandbox-runtime's repository selection to 0.0.79 at d9aac2098351ca17f3743fbaf6ecbd0051b7e00e and synchronize its current saturation-audit identity/receipt references. The sanitized receipt imports the command center's recorded 0.0.79 installation/readback and Claude echo smoke; the existing 0.0.77 filesystem/network evidence keeps its dated scope.

AgentsView changes are owned by #846. At head 20e52faf91601e2b501ad89f4750d684b4ec7502, #846 also carries the plan/display labels, owners mirror, consensus source, generated defaults and handbook chain. All AgentsView edits have been removed from #855 to avoid duplication. The existing owner-pin test is preserved.

• Base commit: e450e378c6bf1b21dd0bb565812f84afe028bad5
• Lane: lane:shared; the other lane's acknowledgement remains a landing requirement.
• Paths: stack/landscape/upstream snapshot, saturation audit, one sanitized SRT receipt, correction log, then publication registry.

SOTA sources

• anthropics/sandbox-runtime v0.0.79 release, published 2026-10-07T17:12:35Z; package.json at d9aac209 and version-specific npm metadata independently identify 0.0.79. Native tag lookup resolves the exact source commit. npm has no gitHead, so exact npm artifact-to-commit provenance remains UNVERIFIED.
• Retained command-center S4 row-70 observation, source receipt SHA256 a35a0577099469de70996dcfa9841a669da20452f1d22432d50df666c7273a61. The imported record cites the pinned upstream README/vendor npm route; no native command was re-executed by this PR.
• The repository's existing unittest modules and native dispatcher follow CPython unittest. These are repository integration tests, not a sandbox-runtime upstream test suite.
• Registry refresh follows docs/lanes.md through the existing registrar and generated-report commands.

Evidence-class table

Limits: the imported source marks Codex's 0.0.79 smoke pending. Historical 0.0.77 policy acceptance is byte-identical and does not qualify those policies at 0.0.79. The receipt distinguishes source-reported command text from the unavailable original Claude post-hook argv. No new installation, native smoke, upstream policy suite, dual-client acceptance or platform-status promotion is claimed.

Hosted/local comparison

The failing hosted run 37759591614 used Ubuntu 24.04 image 20261004.327.1 and logged CPython 3.12.3, with unittest discovery split into eight shards. Local python3 is 3.13.16. All original three symptoms reproduced locally: the earlier clean validate.py result covered publication structure but missed the identity and owner-contract tests.

The first broader local run is retained: 10610 tests, 80 failures, 18 errors, 905 skips, rc1. It exposed 64 line-anchor assertions from inserting an evidence-ID line; the new SRT receipt reference now preserves the established stack line anchors. It also exposed the generated AgentsView producer chain, which remains with #846. Missing local exchange_calendars and local fixture/tool environment failures are recorded; no dependency installation was attempted during the quiet window. The final scoped suite is the separate result above, not a claim that this broader environment has been qualified.

Local commands run

These strings come from the native post-hook exec records, through /bin/bash -lc; changed emitted commands would be recorded separately. Native records and failed attempts are retained privately.

python3 -m unittest -v tests.test_stack_lifecycle.StackLifecycleTests.test_every_selected_component_has_current_identity_and_receipts tests.test_new_wsl_client_config.PlanConfigurationTests.test_plan_checker_counts_direct_config_installs_and_refuses_an_uncopied_file tests.test_new_wsl_definitive_defaults.InterimPlanChecks.test_the_unchanged_copy_passes
Before: rc1, 3 failures

env PYTHONFAULTHANDLER=1 nice -n 19 python3 -m unittest -v tests.test_stack_lifecycle tests.test_architecture_pin_source tests.test_ecosystem_manifest tests.test_new_wsl_client_config tests.test_new_wsl_definitive_defaults tests.test_new_wsl_handbook tests.test_blind_checkout
575 tests, 0 failures, 0 errors, 3 skips; rc0

python3 scripts/validate.py
rc0 — 69 components, 10821 hashed files, 4 profiles, 227 receipts; integrity/scope checks only

git diff --check
rc0

Decision record

CC093314Z §3 directs the pin correction; CC101427Z §3 directs the CI repair. The missed identity/owner checks and actual failure are recorded in docs/harness-defaults.md's anti-pattern log. #846's owner confirmed the complete AgentsView overlap, and primary read-only branch inspection corroborated those paths.

Host evidence

No evidence/hosts/ file is changed. The new public receipt strips private paths and records only the retained SRT observation, returned hashes and explicit limits.

Checklist

• Current selection and historical native evidence remain distinct.
• No credentials, personal host paths or raw conversations are published.
• No host/runtime install, gateway restart, wave-1 job or paper-key operation was performed.
• 2604 drift fixes #846's ownership and peer worktrees are preserved; the trading-runtime freeze remains binding.