Buck

Buck is the engineering agent. He handles code planning, implementation, code review, production investigations, and cross-repo work. When a task requires hands-on development, Buck dispatches ephemeral EKS session pods running Claude Code.

Entry points

Slack — mention @buck in any channel where the Slack app is installed
Jira — assign a ticket to Buck or comment on a ticket he is already working on
Dashboard — create sessions directly at bucky.fetchrewards.com
Pull requests — automatic code review on PR open/update, or ad-hoc @bucky comments

Capabilities

MCP tools

Buck connects to external systems through MCP servers:

Sourcegraph — code search across all repositories
Grafana — observability data, dashboards, and alerts

Custom tools

Buck has purpose-built tools for orchestrating work:

create_jira_ticket / search_jira — Jira integration
create_session / run_claude_session / resume_claude_session — session lifecycle management
respond_in_slack / respond_in_jira — cross-platform responses
post_to_feed — activity feed updates
set_team_jira_project — team configuration

EKS session pods

When Buck needs to write code, he dispatches an ephemeral Kubernetes Job. Each pod runs Claude Code with tmux, MCP plugins, and full git/gh access. Pods are cleaned up automatically after completion.

Multi-repo support

Buck natively supports multiple repositories in a single session — useful for changes that span services or shared libraries.

Session lifecycle

Dispatch — Buck creates a Kubernetes Job with the task context
Clone — the pod clones the target repository (or repositories)
Setup hooks — lifecycle hooks are configured for callbacks
Claude Code — Claude runs inside the pod, writing and testing code
Callback — results (plan, PR, investigation) are reported back to the originating platform

Interactive features

Terminal access — while a session is running, connect to the pod’s terminal from the dashboard in real-time. Watch Claude work or interact directly.
Pause and resume — running sessions can be paused. Workspace and conversation state is saved to S3. Reply with follow-up instructions to resume from exactly where it left off.
Real-time status — session status updates are streamed via SSE (pending, running, succeeded, failed).