S00-015: Verify recovered execution path for interrupted review

Changes: Record S00-015 readiness after independently reviewed S00-016 provider recovery; replace blocked handoff with concrete QA next steps. No runner or test source changes; parent acceptance remains separate.
Validation: Four focused automation tests passed in 4.560s; tasks.py validate passed for 281 tasks at revision 32. Read-only assertions verified the original capacity failure, committed successful resolver/QA transitions, fresh shared-model sessions and unaccepted parents. Git diff whitespace and scope checks passed. Future provider capacity and parent implementation acceptance are not guaranteed.

Task: S00-015
Agent: auto-ci-cd-84e209103f24
Role: ci-cd
Action: claim
Status: review

Task title: Устранить прерывание: Fix runner prompt delivery after partial stdin timeout

state/evidence/S00-015-auto-ci-cd-84e209103f24.md

@@ -0,0 +1,31 @@
+# S00-015 interruption recovery evidence
+
+Executor: auto-ci-cd-84e209103f24 (ci-cd), 2026-10-15.
+Result: interrupted review execution path is observed recovered after reviewed S00-016; S00-015 awaits independent QA. No source changes needed.
+
+## Input, output and acceptance
+
+Input: assigned card/packet, required project instructions and artifacts, prior handoff, accepted ADR 0000, direct dependency executor/review evidence and actual run records. Output: one bounded readiness record for retrying the interrupted S00-012 review. Scope remains scripts/runner.py and tests/tooling/test_automation.py; neither was changed. Only task evidence/handoff/commit-message metadata is authored.
+
+Original interruption: state/runs/auto-qa-98b05bd56c25/run.json records returncode 1 and stopped_after_transition=false. Its output.log reports HTTP 503 Service Unavailable and Selected model is at capacity. This persisted failed startup reproduces the reported cause; no fresh capacity failure or provider retry is claimed.
+
+External correction was handled by separate S00-016 and independently accepted. Its executor auto-ci-cd-351a0cbbabef and QA auto-qa-309d45d4fd4b now have finished run records, stopped_after_transition=true, committed transitions and empty scope violations. Their result.status values are review and done respectively. S00-016 is done. This provides actual recovered fresh reviewer startup and tool execution, not merely an executor availability assumption. Coordinator external readiness and release are documented in S00-016 executor/review evidence.
+
+Read-only inspection confirms runner.local.json has no commands_by_role overrides or resume argument. Original failed QA, S00-016 executor, S00-016 QA and this S00-015 executor have four distinct session ids; all startup headers show gpt-6.1-sol and provider openai. This current post-review fresh session successfully executed actual shell/Python tools. No agent/model was launched by this executor and no provider/configuration setting was changed.
+
+Runner still passes persisted UTF-8 prompt file as binary stdin and polls process.wait, retaining transition/deadline monitoring and cleanup. The existing delayed-reader regression verifies the exact entire 140024-byte UTF-8 payload and EOF after a 0.5-second startup delay. Existing real-descendant checks confirm successful submit and timeout prevent a delayed source write. An incomplete-agent-exit check confirms failure still creates remediation rather than false acceptance.
+
+## Actual commands and results
+
+- python3 scripts/tasks.py resume --agent auto-ci-cd-84e209103f24: exit 0, S00-015 already assigned in_progress; no repeat claim.
+- python3 scripts/tasks.py show S00-015 and packet S00-015 --agent auto-ci-cd-84e209103f24: exit 0 each; required reads, bounded scope and current dependencies inspected. All required input artifacts exist and were read.
+- PYTHONPATH=tests/tooling python3 -m unittest test_automation.RunnerTests.test_delayed_reader_receives_complete_large_utf8_prompt_and_eof test_automation.RunnerTests.test_submit_transition_stops_source_writing_descendant test_automation.RunnerTests.test_timeout_stops_source_writing_descendant test_automation.RunnerTests.test_incomplete_agent_exit_creates_remediation -v: exit 0, 4 tests in 4.560s. Isolated temporary fixtures only, no product task/model launch or Git mutation.
+- python3 scripts/tasks.py validate: exit 0, 281 tasks, 22 stages, 221 requirements, 211 allocated Web requirements, revision 32.
+- Inline read-only Python assertion probe over state/tasks.json, runner.local.json and the four named state/runs/<agent>/run.json/output.log records: exit 0. Asserted original returncode/503/capacity failure; common command without resume/role override; distinct same-model session ids; committed finished S00-016 executor and QA transitions with no violations; current running S00-015 assignment and matching LUMA_AGENT_ID; S00-016 done and S00-012/S00-002 blocked. Output: PASS: original capacity failure; committed recovered executor and independent QA; four distinct shared-model sessions; current fresh tool use; parents unaccepted.
+- git status --short initially: exit 0, only coordinator state/tasks.json claim metadata changed. Final git diff --check: exit 0. Final git status --short: exit 0; only state/tasks.json, this evidence, S00-015 handoff and this agent commit message changed. No source file changed.
+
+## Limits and next step
+
+Observed provider recovery does not guarantee future capacity. This is S00-015 readiness work, not an S00-012 implementation review or S00-002 stack acceptance. The original failure was diagnosed from persisted logs; no deliberate live failure/retry was attempted. Current run completion/commit is recorded by the coordinator after submit. No application/browser/server/Windows readiness is asserted; make check was not rerun because this task changes only workflow documentation and four relevant existing regressions plus task validation passed.
+
+Independent QA must check this evidence and actual recovered records, repeat necessary focused checks, and decide only S00-015. After its reviewed done and coordinator commit, a fresh reviewer can resume S00-012 through the normal workflow. No parent status bypass, retired-id reuse, dependencies, secrets, specs or shared contracts changed.