Skip to content

ci: stabilize live harness shards and MCP App E2E policy - #683

Merged
jmagar merged 5 commits into
mainfrom
fix/ci-live-harness-isolation-20260916
Sep 17, 2026
Merged

jmagar merged 5 commits into
mainfrom
fix/ci-live-harness-isolation-20260916

Conversation

@jmagar

@jmagar jmagar commented Sep 16, 2026

Copy link
Copy Markdown
Collaborator

Problem

The hosted integration split exposed two different failures that were being conflated:

  1. copies of the shared live_labby process harness were running concurrently across many integration binaries, exhausting unchanged readiness/cleanup budgets under hosted load; and
  2. Live E2E's MCP App qualification still assumed the manager resource started disabled, while fresh Labby installs now intentionally default the manager UI on.

A policy-only change to .config/nextest.toml also was not guaranteed to route through the full Rust test path.

Change

  • serialize every embedded live_labby::* support case across platforms
  • serialize integration binaries that own real daemon/browser/process state without inflating product deadlines
  • de-duplicate embedded harness tests from labby-int-* shards and run them once through live_process_harness
  • route .config/nextest.toml changes through the full Rust test path and pin the policy in regression tests
  • bound Agent runtime duration to the authority-lease contract instead of the old 24-hour ceiling
  • keep action_matrix_completeness from importing unrelated live-process support
  • align MCP App qualification with the default-on manager policy: disable via the real mcp_app control tool, then prove direct resource reads cannot bypass the disabled policy
  • make the browser qualification wait on the actual manager switch state rather than brittle status text
  • restore scripts/ci/changed_paths.py executable mode

Local verification

  • cargo fmt --all -- --check: pass
  • git diff --check: pass
  • workspace all-features/all-targets Clippy with -D warnings: pass
  • runtime authority-lease bound regression: pass
  • nextest/changed-path policy regressions: 2/2 pass
  • action matrix: 15/15 pass
  • MCP App qualification: 1/1 pass
  • all five labby-int-* shards pass serially with the de-dup policy
  • docs: 17 generated artifacts fresh, 603 local links, 116 canonical docs, 5/5 contract tests
  • full Live E2E produced green receipts for contracts, HTTP/CLI/API, observability, IPv6, MCP parity, and identity/protected/restart; the final MCP App host shard passes locally with Macpoo's Playwright cache. GitHub Actions uses the script's native /home/runner/.cache/ms-playwright path.

Fresh CI on this head is the authoritative hosted-runner proof. Merge only after every hosted lane, including Live E2E core, is green.

@github-actions github-actions Bot added the documentation Improvements or additions to documentation label Sep 17, 2026
@jmagar jmagar changed the title ci: isolate live harness tests across hosted shards ci: stabilize live harness shards and MCP App E2E policy Sep 17, 2026
@jmagar
jmagar merged commit 29ca96e into main Sep 17, 2026
70 checks passed
@jmagar
jmagar deleted the fix/ci-live-harness-isolation-20260916 branch September 17, 2026 14:09
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation Improvements or additions to documentation

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant