Repository navigation
fix(c7): C7 backfill slice C2 -- 11 durable-state findings in cron/plugins/CI scripts (t_779040d0) - #1371
Conversation
FleetReviewReview: pre-merge · head profile: light (rule: default light: lines 375<800, files 17<1000000, hunks 30<1000000, no hot path) · round 0 · members: B-assert-ctx, L6, C-assert-xhigh, F · families: anthropic,openai Confidence: 3/5 Findings
FleetReview provenance · models: B=gpt-6-sol, C=claude-code-opus-5-5, F=gpt-6-sol · cost: $4.80 · duration: 4m 29s · rounds: 1 · files examined: 17 |
f615e89 to
21d18c3
Compare
FleetReviewReview: pre-merge · head
Reviewed with 1 of 2 model families — anthropic unavailable. profile: light (rule: default light: lines 596<800, files 19<1000000, hunks 39<1000000, no hot path) · round 1 · members: B-assert-ctx, L6, C-assert-xhigh, F · families: openai Confidence: 1/5 Findings
FleetReview provenance · models: B=gpt-6-sol, C=claude-code-opus-5-5, F=gpt-6-sol · cost: $7.46 (estimated) · duration: 3m 48s · rounds: 2 · files examined: 19 |
|
🤖 merged-by: apollo · lane: kanban-merge-pass · gate: ADVISORY (FleetReview not green for 21d18c3): fleetreview-advisory-20260927-standing.md · why: t_779040d0: C7 backfill fix: hermes-agent slice C2: cron, plugins, CI scripts (14 instances); Argus off card review (Ace 13:08), CI green |
788221e to
ef35814
Compare
FleetReviewReview: pre-merge · head
Reviewed with 1 of 2 model families — anthropic unavailable. profile: light (rule: default light: lines 596<800, files 19<1000000, hunks 39<1000000, no hot path) · round 1 · members: B-assert-ctx, L6, C-assert-xhigh, F · families: openai Confidence: 1/5 Findings
FleetReview provenance · models: B=gpt-6-sol, C=claude-code-opus-5-5, F=gpt-6-sol · cost: $4.78 · duration: 4m 37s · rounds: 2 · files examined: 19 |
ef35814 to
137e189
Compare
FleetReviewFleetReview's daily member-call budget is spent (1082/1200 for 2026-09-28 UTC); review skipped. FleetReview · reviewKind: skipped-budget |
- cron/scheduler.py: flush host-down ledger rows the moment the #logs send succeeds, not after all targets. - mem0 capture_router: match staged turn file literally (glob.escape). - ci_overflow_integration full_rerun_verdict: never-started jobs are not reused/missing; a prior-attempt job absent from the re-run BLOCKs. - blackbox store/_measured + prefix guard: gate each bucket on its own unknown flag; aggregate-only usage_unknown keeps measured buckets. Six new tests, each RED on the pre-fix head, GREEN after.
FleetReviewFleetReview's daily member-call budget is spent (1082/1200 for 2026-09-28 UTC); review skipped. FleetReview · reviewKind: skipped-budget |
…ugins/CI scripts (t_779040d0) Re-verified 14 C7 instances at 533612e (then rebased on ab6b6d4). Every fix has a RED-on-base test. Fixed: - k87 cron/fork_ext/scheduler_ext.py: timeout_s "inf"/float('inf') raised OverflowError out of resolve_job_script_timeout; now falls back. - k88 cron/scheduler.py: host-down ledger row was written before the send; now written only once the demoted #logs delivery succeeds. - k116 plugins/blackbox/__init__.py: prefix guard got cache_read / prompt_tokens as measured numbers when buckets were unknown; now None. - k117 plugins/blackbox/store.py: turn_api_calls stored unknown buckets as 0, so first_call_cache_miss judged an unmeasured read; now NULL. - k119 lcm/db_bootstrap.py: a trigger-only repair cleared a deep/parity corruption flag; now only a rebuild clears it (a structural flag is cleared when the structure verifies). - k120 lcm/db_bootstrap.py: repair rolled back a caller-owned transaction on failure; now rolls back only its own (captured at entry). - k122 mem0/capture_drain.py: the exactly-once shortcut (add committed, crash before routing) skipped Arm-B routing; now routes (staging is keyed by turn id, so a re-route overwrites in place). - k125 scripts/ci/live_comment.py: re-run dedupe kept the first-listed artifact; now newest (created_at, id) wins. - k127 scripts/ci_overflow_integration.py: an executed job with no PROBE line gave PASS; now UNVERIFIABLE. - k128 scripts/ci_overflow_integration.py: one matching slice certified a selective re-run as full; full_rerun_verdict now BLOCKs when any job was carried over from the prior attempt. - k136 plugins/blackbox/store.py: migration test passed on a table that could not take an insert; the migration now adds every base column insert_api_call names, and the test inserts after migrating. Dropped: - k124 FIXED on main by #1353 (live_comment.py:659 `body = last_body`, test_comment_lookup_outage_is_retried_not_fatal). - k126 no rate-limit bug to fix: nothing on main runs live_comment.py (no .github reference; ci-review-comment was retired in aff7aea). - k133 FALSE: the autouse tests/conftest.py::_hermetic_environment sets HERMES_HOME per test; a probe confirmed get_hermes_home() is <tmp_path>/hermes_test, so the real scripts dir is never written.
- cron/scheduler.py: flush host-down ledger rows the moment the #logs send succeeds, not after all targets. - mem0 capture_router: match staged turn file literally (glob.escape). - ci_overflow_integration full_rerun_verdict: never-started jobs are not reused/missing; a prior-attempt job absent from the re-run BLOCKs. - blackbox store/_measured + prefix guard: gate each bucket on its own unknown flag; aggregate-only usage_unknown keeps measured buckets. Six new tests, each RED on the pre-fix head, GREEN after.
550a9a9 to
3e8ef74
Compare
FleetReviewFleetReview's daily member-call budget is spent (1082/1200 for 2026-09-28 UTC); review skipped. FleetReview · reviewKind: skipped-budget |
FleetReviewReview: pre-merge · head profile: light (rule: default light: lines 693<800, files 19<1000000, hunks 38<1000000, no hot path) · round 2 · members: B-assert-ctx, L6, C-assert-xhigh, F · families: anthropic,openai Confidence: 3/5 Findings
FleetReview provenance · models: B=gpt-6-sol, C=claude-code-opus-5-5, F=gpt-6-sol · cost: $5.64 · duration: 6m 56s · rounds: 1 · files examined: 19 |
|
🤖 merged-by: apollo · lane: review · gate: ADVISORY (FleetReview not green for 430c5ba): fleetreview-advisory-20260927-standing.md · why: t_779040d0 C7 slice C2 rebased 3x; CI 24/24; armed but never queued |
1 participant
Re-verified 14 C7 instances at 533612e (then rebased on ab6b6d4).
Every fix has a RED-on-base test.
Fixed:
OverflowError out of resolve_job_script_timeout; now falls back.
now written only once the demoted #logs delivery succeeds.
prompt_tokens as measured numbers when buckets were unknown; now None.
0, so first_call_cache_miss judged an unmeasured read; now NULL.
corruption flag; now only a rebuild clears it (a structural flag is
cleared when the structure verifies).
on failure; now rolls back only its own (captured at entry).
crash before routing) skipped Arm-B routing; now routes (staging is
keyed by turn id, so a re-route overwrites in place).
artifact; now newest (created_at, id) wins.
line gave PASS; now UNVERIFIABLE.
selective re-run as full; full_rerun_verdict now BLOCKs when any job
was carried over from the prior attempt.
could not take an insert; the migration now adds every base column
insert_api_call names, and the test inserts after migrating.
Dropped:
body = last_body,test_comment_lookup_outage_is_retried_not_fatal).
.github reference; ci-review-comment was retired in aff7aea).
HERMES_HOME per test; a probe confirmed get_hermes_home() is
<tmp_path>/hermes_test, so the real scripts dir is never written.
Need help on this PR? Tag
@codesmith-botwith what you need. Autofix is disabled.