[scott-lowe-vapi]

Support partial GitOps promotions using traffic splitting

Part of DEPLOY-205. 5 of 5: make a stuck ramp visible, and count health-gate aborts apart from aborts someone ran.

What changes

• Health-gate aborts are written as rollout-health-abort instead of rollout-abort. That is one more value in RolloutCommand, and no server change, since the x-vapi-client schema accepts [a-z0-9_-]. PostHog can now count how often the gate fires.
• Each advance and status run appends a table to $GITHUB_STEP_SUMMARY: every ramp's state and what the health gate read (calls and error rate per version). Before, a stuck ramp was a console.log line in a green run.
• advance fails when a ramp is overdue, so GitHub's failure notification reaches someone. Overdue means:
  • a step still waiting three intervals after it was written (the gate hasn't seen minCalls);
  • a stale pin from a killed promotion, left for a human after 6h;
  • the split changed under a step. Only the run that finds it fails, since the next tick no longer sees a ramp.
• Holds and aborts are deliberate, so they never fail a run. Failing on them would turn every 15-minute tick red until the next promote.
• status lists overdue ramps but doesn't fail.
• Docs: promotion.md covers the summary, what fails a run, and who GitHub notifies; how-it-works.md lists the new command.

This doesn't catch a ticker that never runs: a disabled schedule, an Actions budget block, or VAPI_ROLLOUT_ENABLED unset. That needs the server-side stuck-ramp insight in PostHog, which waits on GitOps attribution reaching production.

Evidence

• rollout-tick.test.ts: the gate abort's command; overdue for a starved gate after three intervals, a stale pin, and a malformed step; never overdue for a hold or an abort; the gate reading on a passing step.
• rollout-cmd.test.ts: a split changed under a step fails that run, and the next one passes; a stale pin fails advance while a hold beside it doesn't; status lists it without failing; the exact step summary.
• rollout-summary.test.ts: the table, its overdue count, pipe escaping, and the failure message.

Tests run with GITHUB_STEP_SUMMARY unset unless a test sets it, so CI's own summary stays clean. 695 tests pass (682 on #89).

🤖 Generated with Claude Code

[scott-lowe-vapi]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• feat(rollout): report each tick's ramps and fail the run when one is overdue #90 👈 (View in Graphite)
• feat(rollout): gate each ramp step on the new version's calls and errors #89
• feat(rollout): add the rollout ticker, its workflow and promotion's rollout input #88
• feat(promote): pin assistants around the apply so a rollout org ramps instead of cutting over #87
• feat(promote): add rollout config, traffic-allocation client and splitting probe #86
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.