team-upgrade

Rolls out new agent image versions to a team using staged canary upgrades with rollback.

2|Updated Apr 5, 2026
One-click install
npx skills add https://github.com/witwave-ai/witwave --skill team-upgrade-witwave-ai
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: team-upgrade
Source: https://github.com/witwave-ai/witwave/tree/main/.agents/self/milo/.claude/skills/team-upgrade
Command: npx skills add https://github.com/witwave-ai/witwave --skill team-upgrade-witwave-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Upgrading a fleet of AI agents to a new release risks breaking the whole team at once. This Skill performs a staged, self-canary rollout: the operator agent upgrades itself first, soaks on the new version, then cascades to peers one at a time, rolling back and quarantining any bad version before it spreads. ## Core Features & Use Cases - Self-canary rollout: The agent upgrades itself first, soaks for a configurable window (default 24h), and only cascades to peers once proven healthy. - Ordered peer cascade: Peers upgrade one per run in lowest-blast-radius order, with health verification and automatic rollback to last-known-good on failure. - Campaign state management: Pinned campaign targets, quarantine lists, kill-switch pause, and an append-only audit log persisted in memory across runs spanning hours to days. - Use Case: A new Witwave release is published. The skill detects the drift, upgrades the Milo agent first, verifies health over a soak period, then rolls the version to piper, mira, and the remaining agents one at a time via ww agent upgrade. ## Quick Start Ask the agent to check for updates and roll out the latest released version to the team safely.

Frequently Asked Questions about team-upgrade

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I safely upgrade a team of AI agents to a new version?▼

Use a staged canary rollout: upgrade one agent first, let it soak on the new version, then cascade to peers one at a time in lowest-blast-radius order. This Skill automates that flow through the ww CLI with verification and rollback at each step.

How does the canary upgrade and rollback process work?▼

The operator agent upgrades itself first via ww agent upgrade, records last-known-good tags, and soaks for a configurable window (default 24h). If health checks fail, it rolls itself back, quarantines the bad version, and halts the campaign before any peer is touched.

Can I pause or stop an in-progress team upgrade campaign?▼

Yes. Creating a PAUSED file in the skill's memory directory acts as a kill switch; every subsequent run reports paused and takes no action. Quarantining the in-flight target version also halts the campaign.

What happens if an agent upgrade fails during the rollout?▼

The failed agent is rolled back to its last-known-good tags via ww agent upgrade, the target version is added to quarantine.md, the campaign target is cleared, and the incident is escalated. That version is never retried; the team waits for a newer release.

Does this skill upgrade the Kubernetes operator or agent model config?▼

No. It only upgrades agent images (harness plus backends together to one tag). Operator upgrades require cluster-admin and are escalated to a human, and model or effort config lives outside image tags, so it is recommended to humans rather than pushed.