coulomb/ops-warden
5
0
Fork 0
Code Issues Pull requests Projects Releases 2 Packages Wiki Activity Actions

test: isolate ordinary unit runs from ambient credentials and provider I/O
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
Details
CI Smoke / container-smoke (push) Successful in 1s
Details

Browse source 
Assistant: codex
Assistant-Model: gpt-6.1-sol
Assistant-Session: 01a1468b-63fa-7710-85dc-8f2e16b3ebfa
This commit is contained in:
tegwick 2026-10-17 21:30:12 +02:00
parent 922c26da0f
commit b593c87983
14 changed files with 592 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

38
workplans/WARDEN-WP-0044-security-case-follow-through.md
View file

@ -354,3 +354,41 @@ then use the preserved isolated review runner, not the ambient `make test` path.
No credential lookup, live investigation or rotation is authorized by this note.
T03 remains progress; contained sessions, authenticated owner provenance,
materialization, exact type acknowledgment/pin and first native case remain open.
2026-10-17 safe unit-runner source candidate: `make test` now selects an existing
interpreter and `scripts/run_unit_tests.py`, which constructs the environment and
empty temporary HOME before pytest collection, denies unexpected process/socket
use and ambient file opens, and guards the pinned BrowserInputs child too.
Closed fixture exceptions cover the synthetic desk endpoint, metadata checkers,
exact-byte contract doubles and fixed transport prints; they confer no provider
authority. Explicit public declaration fixtures replace ambient HOME source
lookups. See `docs/development/safe-unit-tests.md` and Custodian packet
`docs/changes/netkingdom-target-2026-10-17/warden-safe-unit-runner-r1`.
Synthetic token, ambient-path, forged-provider and nested-child regressions are
retained alongside the initial failed isolation runs and corrected results. This
is a source candidate for the existing safe-test handoff, not an OS sandbox or
native case acceptance. T03 remains progress; contained sessions, authenticated
owner provenance, materialization, exact type acknowledgment/pin and first native
case remain open. No credential lookup, live investigation or rotation occurred.
Candidate validation: ordinary `make test` under synthetic ambient token/HOME
controls passes 783 tests with four integration-marked tests deselected. Affected
focused checks pass 117 tests; new runner code/regressions pass Ruff. Retained
initial failures exposed fixture dependencies and were corrected, not suppressed.
2026-10-17 root follow-through: the original safe-runner source candidate was
independently rejected for O_TRUNC open-write misclassification and unchecked
public fixture projection hashes. Both counterexamples are retained. The minimal
correction treats truncating/appending opens as writes and adds the closed
projection-byte provenance regression. Root inspected the two exact corrections
and ran ordinary make test on stable corrected source with only synthetic parent
HOME/token/Python/pytest sentinels: 785 passed, four deselected. No real credential
or native provider action occurred. Initial author failures and rejected review
are preserved in the Custodian warden-safe-unit-runner-r1 and
warden-safe-unit-runner-review-r1 packets, with root disposition/full result in
warden-safe-unit-integration-r1. This completes the known trusted-unit-entry
accident-prevention follow-through; direct pytest bypass, native/non-open/preopened
file-descriptor limits remain explicit. No OS sandbox, isolated agent assignment,
production grant or native T03 acceptance follows. T03 stays progress.