Repository navigation
feat(alerts): investigate platform-owned insight firings with the agent - #127923
trunk-io[bot] merged 8 commits into
Conversation
A firing of an alert the platform owns now follows the insight stack's investigation rules. The check decides whether the firing gets an investigation, holds a gated alert's delivery until the verdict, and mutes a firing past the budget whose last verdict suppresses it. Only an owned check decides, so a parallel run never pays for an agent run. After the record, a new claim activity writes the investigation row and the evaluate workflow starts the agent on the AI queue, behind a patch. A refused claim releases the held firing by the firing's last verdict. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
🤖 CI report
|
| File | Comment lines | Added lines |
|---|---|---|
products/alerts/backend/temporal/platform_evaluate.py |
5 | 62 |
products/alerts/backend/platform_investigation.py |
3 | 122 |
products/alerts/backend/platform_investigation_detail.py |
3 | 49 |
products/alerts/backend/platform_source_cycle.py |
2 | 9 |
products/alerts_platform/backend/logic/delivery_holds.py |
2 | 40 |
products/alerts_platform/backend/tests/test_delivery_evaluation.py |
1 | 54 |
This check does not block merging. It updates on every push and clears when the share drops.
⚠️ Backend coverage — 87.0% of changed backend lines covered — 46 uncovered
🧪 Backend test coverage
Patch coverage — changed backend lines (products + core): █████████████████░░░ 87.0% (311 / 357)
| File | Patch | Uncovered changed lines |
|---|---|---|
products/alerts/backend/temporal/platform_evaluate.py |
48.6% | 174–175, 358, 360, 434, 785, 788–790, 795–796, 803, 806–807, 810–811, 818, 820 |
products/alerts_platform/backend/facade/api.py |
50.0% | 89, 100 |
products/alerts/backend/platform_investigation_detail.py |
71.9% | 23–27, 32, 37, 54–55 |
products/alerts_platform/backend/facade/testing.py |
72.7% | 82, 105, 110 |
products/alerts/backend/facade/investigations.py |
73.3% | 23, 28, 33, 46 |
products/alerts/backend/investigation_release.py |
88.9% | 95–96, 99, 105–106, 109 |
products/alerts/backend/platform_investigation.py |
92.5% | 157, 197, 201, 227 |
🤖 Agents: add a test only if an uncovered line exposes a realistic regression that existing tests miss. Otherwise explain why no new test is needed under "How did you test this code?". Gap list: the patch-coverage artifact on this run (gh run download 355140927939397 -n patch-coverage), or the coverage-data block at the end of this comment.
Per-product line coverage (touched products)
| Product | Coverage | Lines |
|---|---|---|
platform_features |
██░░░░░░░░░░░░░░░░░░ 12.1% |
7 / 58 |
data_tools |
████████████░░░░░░░░ 60.9% |
95 / 156 |
warehouse_sources_queue |
█████████████░░░░░░░ 65.2% |
2,161 / 3,314 |
demo |
█████████████░░░░░░░ 65.7% |
1,803 / 2,743 |
batch_exports |
████████████████░░░░ 82.2% |
23,567 / 28,654 |
engineering_analytics |
█████████████████░░░ 84.9% |
11,699 / 13,779 |
commands |
█████████████████░░░ 85.4% |
1,621 / 1,898 |
ml_inference |
█████████████████░░░ 86.3% |
524 / 607 |
mcp_analytics |
█████████████████░░░ 86.3% |
5,803 / 6,722 |
posthog_ai |
██████████████████░░ 87.9% |
4,303 / 4,895 |
apm |
██████████████████░░ 88.2% |
1,973 / 2,236 |
notebooks |
██████████████████░░ 88.5% |
15,356 / 17,347 |
mcp_registry |
██████████████████░░ 89.0% |
1,596 / 1,794 |
webmcp |
██████████████████░░ 89.3% |
292 / 327 |
dashboards |
██████████████████░░ 89.4% |
8,048 / 9,003 |
product_tours |
██████████████████░░ 89.7% |
1,355 / 1,510 |
warehouse_suggestions |
██████████████████░░ 89.9% |
3,707 / 4,124 |
cohorts |
██████████████████░░ 90.1% |
8,996 / 9,990 |
signals |
██████████████████░░ 90.5% |
99,655 / 110,121 |
today |
██████████████████░░ 90.6% |
3,495 / 3,859 |
visual_review |
██████████████████░░ 90.7% |
9,852 / 10,863 |
managed_warehouse |
██████████████████░░ 91.2% |
12,686 / 13,914 |
canvas |
██████████████████░░ 91.2% |
8,203 / 8,994 |
tasks |
██████████████████░░ 91.3% |
82,102 / 89,969 |
data_warehouse |
██████████████████░░ 91.3% |
15,700 / 17,201 |
autoresearch |
██████████████████░░ 91.5% |
11,717 / 12,805 |
data_modeling |
██████████████████░░ 91.5% |
11,198 / 12,233 |
cdp |
██████████████████░░ 91.5% |
10,432 / 11,396 |
metrics |
██████████████████░░ 91.7% |
11,234 / 12,253 |
slack_app |
██████████████████░░ 91.8% |
21,946 / 23,913 |
wizard |
██████████████████░░ 91.8% |
6,868 / 7,479 |
streamlit_apps |
██████████████████░░ 91.9% |
3,139 / 3,415 |
business_knowledge |
██████████████████░░ 92.0% |
10,505 / 11,423 |
review_hog |
██████████████████░░ 92.3% |
17,768 / 19,256 |
conversations |
██████████████████░░ 92.3% |
31,158 / 33,746 |
error_tracking |
███████████████████░ 92.5% |
20,487 / 22,146 |
authorization |
███████████████████░ 92.5% |
544 / 588 |
alerts_platform |
███████████████████░ 92.6% |
8,487 / 9,170 |
stamphog |
███████████████████░ 92.6% |
8,871 / 9,584 |
early_access_features |
███████████████████░ 92.6% |
1,339 / 1,446 |
notifications |
███████████████████░ 92.8% |
1,218 / 1,312 |
alerts |
███████████████████░ 92.9% |
11,312 / 12,176 |
aeo |
███████████████████░ 93.0% |
2,150 / 2,312 |
surveys |
███████████████████░ 93.3% |
7,248 / 7,772 |
ai_training |
███████████████████░ 93.3% |
598 / 641 |
web_analytics |
███████████████████░ 93.4% |
26,448 / 28,325 |
ai_gateway |
███████████████████░ 93.6% |
787 / 841 |
messaging |
███████████████████░ 93.7% |
6,756 / 7,214 |
approvals |
███████████████████░ 93.7% |
5,004 / 5,341 |
cross_project_dashboards |
███████████████████░ 93.7% |
862 / 920 |
marketing_analytics |
███████████████████░ 93.7% |
21,955 / 23,420 |
context_layer |
███████████████████░ 93.8% |
3,984 / 4,247 |
exports |
███████████████████░ 93.8% |
4,314 / 4,597 |
workflows |
███████████████████░ 93.8% |
27,397 / 29,194 |
mcp_store |
███████████████████░ 93.9% |
9,188 / 9,784 |
ai_observability |
███████████████████░ 93.9% |
32,676 / 34,793 |
billing_alerts |
███████████████████░ 94.0% |
2,322 / 2,470 |
customer_analytics |
███████████████████░ 94.0% |
29,900 / 31,794 |
managed_migrations |
███████████████████░ 94.1% |
1,877 / 1,994 |
replay_vision |
███████████████████░ 94.4% |
33,326 / 35,317 |
legal_documents |
███████████████████░ 94.4% |
2,313 / 2,451 |
logs |
███████████████████░ 94.5% |
19,432 / 20,567 |
experiments |
███████████████████░ 94.7% |
43,959 / 46,433 |
reminders |
███████████████████░ 94.8% |
782 / 825 |
actions |
███████████████████░ 94.9% |
997 / 1,051 |
endpoints |
███████████████████░ 94.9% |
9,566 / 10,077 |
growth |
███████████████████░ 95.0% |
11,563 / 12,176 |
annotations |
███████████████████░ 95.1% |
1,097 / 1,153 |
skills |
███████████████████░ 95.2% |
7,633 / 8,018 |
uptime_monitoring |
███████████████████░ 95.3% |
306 / 321 |
access_control |
███████████████████░ 95.5% |
8,358 / 8,753 |
subscriptions |
███████████████████░ 95.6% |
13,645 / 14,275 |
data_catalog |
███████████████████░ 95.7% |
4,715 / 4,928 |
product_analytics |
███████████████████░ 95.8% |
30,289 / 31,620 |
feature_flags |
███████████████████░ 95.8% |
29,654 / 30,953 |
revenue_analytics |
███████████████████░ 95.9% |
1,877 / 1,958 |
data_quality |
███████████████████░ 96.0% |
8,561 / 8,915 |
tracing |
███████████████████░ 96.5% |
4,564 / 4,729 |
warehouse_sources |
███████████████████░ 96.5% |
492,991 / 510,624 |
account_audits |
███████████████████░ 96.7% |
770 / 796 |
security |
███████████████████░ 96.8% |
1,597 / 1,650 |
pulse |
███████████████████░ 97.3% |
2,036 / 2,093 |
analytics_platform |
████████████████████ 97.9% |
2,862 / 2,922 |
groups |
████████████████████ 98.3% |
1,864 / 1,897 |
field_notes |
████████████████████ 99.4% |
172 / 173 |
Report-only. Patch coverage = changed backend lines covered vs origin/master. Sorted lowest first.
Known gaps: lines covered only by Temporal tests show as uncovered; core line numbers may drift if master changed the same file.
![posthog[bot]](https://avatars.githubusercontent.com/in/1203049?s=60&v=4){kind=small}
There was a problem hiding this comment.
PostHog Review · Standard · reviewed bf93cb0: 2 Should fix.
This is the only automatic review while this pull request is a draft. Comment @posthog review to review new changes sooner. Otherwise, a review of the new changes runs when you mark it ready for review.
See this review and its findings in PostHog Review. Comment @posthog review --deep for a Deep review. A Deep review takes longer and looks deeper than a Standard review.
…ght-investigation-adapter
…ght-investigation-adapter
|
Risk: No findings This delta fixes the timezone differential in the hold-release staleness check by normalizing both the Postgres firing start and the ClickHouse episode start to aware UTC datetimes before comparing, and adds a regression test that feeds a zoneless episode start through the real comparison path. No new behavior or security surface is introduced beyond the comparison helper. Sentinel reviewed |
…ght-investigation-adapter # Conflicts: # products/alerts/backend/platform_investigation.py
… record loop A skipped check of a firing alert keeps its state without evaluating, so it no longer asks for an agent run. Investigation claims now run as tasks after each chunk's record, awaited before the run returns, so a slow claim no longer holds back the next record or admission round. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ght-investigation-adapter
…has ended A held evaluation can be released after a later one resolved the firing it held, and nothing orders the two sends, so its trigger opened an incident no resolve followed. Delivery now sends a released hold's incident trigger only while its group is still in the firing the held evaluation saw. The message still goes out. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
![stamphog[bot]](https://avatars.githubusercontent.com/in/3000332?s=60&v=4){kind=small}
There was a problem hiding this comment.
Not approved — this change needs a human reviewer.
Re-add the stamphog label to request another review once you have addressed this.
Owning-team authorship and STRONG familiarity provide assurance for the risky write path, but @PostHog's substantive incident-ordering concern remains unresolved. Address that thread with episode-aware delivery coordination and get @PostHog/team-apm to review the fix before re-requesting.
- Author wrote 100% of the modified lines and has 25 merged PRs in these paths (familiarity STRONG).
- @PostHog, products/alerts/backend/platform_investigation.py: the held-trigger versus recovery-resolve ordering concern remains unfixed. with_investigation holds the original delivery, while platform_source_cycle.py::_incident_actions independently generates the recovery resolve; delivery_holds.py returns the stored trigger on release without checking whether its episode recovered. Coordinate the held trigger and its episode's resolve, add a recovery-before-verdict regression test, and request review from @PostHog/team-apm before re-requesting.
Gate mechanics and policy version
| Gate | Result | |
|---|---|---|
| prerequisites | ✓ | all clear |
| deny-list | ✓ | no deny categories matched |
| size | ✓ | 276L, 4F substantive, 404L/5F incl. docs/generated/snapshots — within ceiling |
| tier | ✓ | T1-agent / T1d-complex (404L, 5F, single-area, feat) |
| stamphog 2.7.0 | .stamphog/policy.yml @ 7d60245 · reviewed head 7d60245 |
![parameterai[bot]](https://avatars.githubusercontent.com/in/2665233?s=60&v=4){kind=small}
There was a problem hiding this comment.
Not approved — this change needs a human reviewer.
Re-add the stamphog label to request another review once you have addressed this.
Owning-team authorship, strong familiarity, and current-head assurance cover the risky changes, but @parameterai's substantive concern remains unfixed. Address the timestamp comparison in delivery_holds.py and its regression coverage, then request review from @PostHog/team-apm before re-requesting.
- Author wrote 100% of the modified lines and has 42 merged PRs in these paths (familiarity STRONG).
- PostHog Review (Standard) reviewed the current head: no findings.
- @parameterai on products/alerts_platform/backend/logic/delivery_holds.py:183–212, reiterated by @parameterai[bot]: the current staleness comparison still compares timezone-aware PostgreSQL firing timestamps against unnormalized ClickHouse episode timestamps. The announcement reader passes those timestamps through unchanged, while delivery/dispatch.py explicitly normalizes naive ClickHouse timestamps elsewhere. Consequently, a still-current held firing can lose its incident trigger in delivery/evaluation.py, silently preventing PagerDuty notification. Normalize timestamps before comparison and add a test using a naive announcement timestamp before re-requesting.
Gate mechanics and policy version
| Gate | Result | |
|---|---|---|
| prerequisites | ✓ | all clear |
| deny-list | ✓ | no deny categories matched |
| size | ✓ | 325L, 6F substantive, 499L/8F incl. docs/generated/snapshots — within ceiling |
| tier | ✓ | T1-agent / T1d-complex (499L, 8F, single-area, feat) |
| stamphog 2.7.0 | .stamphog/policy.yml @ 2775218 · reviewed head 2775218 |
ClickHouse can return a history row's episode start without a zone, and a naive time never equals the aware one Postgres stores, so every held trigger looked stale and was dropped. Both sides are now compared in UTC. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
Approved.
The changes enter investigation and delivery write paths, with assurance from the owning-team author, STRONG familiarity, and a current-head review reporting no findings. No concrete unresolved review concern or other refusal ground is present.
- Author wrote 100% of the modified lines and has 42 merged PRs in these paths (familiarity STRONG).
- PostHog Review (Standard) reviewed the current head: no findings.
Gate mechanics and policy version
| Gate | Result | |
|---|---|---|
| prerequisites | ✓ | all clear |
| deny-list | ✓ | no deny categories matched |
| size | ✓ | 335L, 6F substantive, 522L/8F incl. docs/generated/snapshots — within ceiling |
| tier | ✓ | T1-agent / T1d-complex (522L, 8F, single-area, feat) |
| stamphog 2.7.0 | .stamphog/policy.yml @ c10005e · reviewed head c10005e |
|
This pull request was merged into |
Problem
Changes
ABANDON, behind a patch. The task never blocks the next record or admission round.How did you test this code?
Test rationale: the source-cycle cases catch a parallel run asking for an agent, a gated firing sent early, a muted firing still delivered, and a refused claim that never releases its hold. They extend the existing owned-check tests.
Ran the alerts and alerts platform suites locally, and repo-wide mypy. Not run: a workflow run against a Temporal server.
Release status
Docs update
🤖 Agent context
Autonomy: Human-driven (agent-assisted)
Agent: Claude Code, Claude Opus 5.5 (
claude-opus-5-5)Skills:
/writing-tests,/writing-code-comments,/writing-user-facing-copy,/stacking-prs,/reviewing-with-coderabbit,/simplify,/code-review,/writing-pr-descriptions🤖 Generated with Claude Code