chore: refresh connections l0 baseline capture dates

Merged
KN
4d ago
+29-29

Problem / Motivation

Goal: Restore a green nightly Connections L0 Static Metadata Probe run on main without weakening the staleness gate.

The scheduled run https://github.com/knowlet/KiroCrew/actions/runs/43604135235 reported failed_count: 0 and all providers ok, but still exited 1. Job logs include STALE BASELINES: {'dropbox': '2026-09-08', 'huggingface': '2026-09-10'} and the report’s staleness.enforced=true block matches the workflow’s --fail-on-stale flag.

Why it matters

The nightly job is the only place --fail-on-stale is enforced (see .github/workflows/connections-l0.yml runbook and l0_probe._staleness). When baselines pass the 60-day L0_VERIFICATION_WARN_AGE_DAYS threshold, the workflow is meant to go red so someone re-records from live providers—not so maintainers learn to ignore a failing schedule.

Not a goal

  • Weakening or removing the staleness gate.
  • Changing scheduled-failure-watch.yml (separate owner decision).
  • Fixing unrelated flaky CI on this fork (see CI failure attribution below).
  • Fixing unrelated local-gate failures in test_registry_store_art_prewarm.py.

What changed (motivation → approach → change)

Symptom: nightly exit 1 with empty drift failures but nonempty staleness.overdue.

Root cause: dropbox and huggingface l0_expectations.verified_on were older than 60 days; on 2026-11-14 the probe sets ok=False when --fail-on-stale and overdue is nonempty (see l0_probe.main around the staleness block).

Approach: run the documented recorder on a networked host: python3 -m kiro_crew.connections.l0_probe --record, which reached 29/29 providers and updated verified_on (and confirmed observed dcr/pkce unchanged).

Change: commit the resulting src/kiro_crew/connections/registry.json diff only.

Backwards compatibility

Compatible: registry schema and probe semantics are unchanged; only baseline capture dates (and implicit re-confirmation of observed flags) were refreshed.

Removes nothing: metadata-only refresh of existing provider entries.

Tests

  • python3 -m pytest -n0 test/test_connections_l0_probe.py test/test_connections_registry.py → 232 passed
  • python3 -m kiro_crew.connections.l0_probe --fail-on-stale locally → exit 0, staleness.overdue: {}

Manual verification

Ran python3 -m kiro_crew.connections.l0_probe --record (29/29 reached, exit 0), then python3 -m kiro_crew.connections.l0_probe --fail-on-stale (exit 0). Reviewed git diff for registry.json: only l0_expectations.verified_on moved to 2026-11-14 for all refreshed providers; no issuer hand-edits.

Screenshots / video

N/A — no user-visible UI change.

Related Issues

Sibling sweep

Pattern harvest

CI failure attribution (scope boundary)

Fork main is pinned at ce07caac5a (2026-10-31). Upstream kirodotdev/KiroCrew main is ahead (d2f22f259e, 2026-11-16); this PR intentionally targets the fork base only (registry.json on that tip).

Comparison method: for each red check on the latest PR workflow run, the failing test or step text was matched against this diff (only registry.json dates). Where knowlet/main has no recent successful full CI matrix, the same failure signatures are treated as pre-existing on the fork base, not introduced by this commit.

CheckPR root causeCaused by this PR?Main (ce07caac) comparison
Coverage GateAggregator: backend-test=failure from unrelated shard redsNoNo comparable green matrix on recent main push
Backend Tests (3.12, 4)test_session_markdown.py::test_a_condition_one_opener_is_closed_as_the_parser_reads_it[long-s]NoSame test file on base; not touched by diff
Backend Tests (3.12, 7)test_terminal_handler_pty_integration.py (2× PROMPT_COMMAND integration tests)NoNot touched by diff
Backend Tests (Windows) (4)Same test_session_markdown case as Linux shard 4NoNot touched by diff
E2E (stub ACP backend, offline)Playwright does not support chromium on ubuntu26.04-x64 (ubuntu-latest runner image)NoRunner-image / Playwright support issue
Pod Boot Canary (Windows)Required runner group 'kirocrew-runners-windows' not foundNoFork infra: private runner group unavailable
GPT 6.1 ReviewJob completes with empty steps (~0s); runner group kirocrew-runners not available on this forkNoSame pattern on prior PR heads (84537f9d…)

In scope for this PR (fixed on head 4931ee92c7): commit metadata — author knowlet, DCO Signed-off-by only, no Co-authored-by; PR description stripped of third-party bot appendages and agent footer links.

Out of scope (pre-existing / fork infra): review lanes (Design/UX/Opus/First Principles), Dependency License Gate, internal-content-scan, and PR Readiness aggregation over the above — none are triggered or fixable by a metadata-only registry.json refresh without widening the PR.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

Contribution License Agreement

Contributions are accepted under the project's Apache-2.0 license; see CONTRIBUTING.md.

Open in Web Open in Cursor 

Summary by cubic

Refreshes the Connections L0 baseline capture dates in registry.json so the nightly probe's --fail-on-stale gate stops failing after dropbox and huggingface exceeded the 60-day warn threshold.

  • Re-records verified_on to 2026-11-14 for all 29 providers; observed dcr/pkce flags are unchanged.
  • Applies only to src/kiro_crew/connections/registry.json; registry schema and probe semantics are untouched.

Summary by Sourcery

Refresh the Connections L0 provider baselines to restore successful nightly staleness validation.

Bug Fixes:

  • Refresh stale Connections L0 baseline capture dates so the nightly staleness check passes without changing probe behavior.

Chores:

  • Update existing provider metadata in the connection registry with newly verified baseline dates.

Summary by CodeRabbit

  • Updates
    • Connection verification dates have been refreshed to November 14, 2026, across the registry.
    • Notion’s manual revocation guidance remains available in its connection information.