AI Dev Workflow (Claude Code)

This site documents how we work with Claude Code in this workspace -- the setup, the task-delivery pipeline, every slash command, the skills and agents, and the rules that auto-apply on every session.

The workspace is an organization folder: each top-level directory is an independent repository. A workspace CLAUDE.md holds the shared conventions and loads a set of rules; each repo may add a supplementary CLAUDE.md.

Start here

If you want to...	Read
Understand the folder layout, CLAUDE.md, and the .claude/ tree	Setup and layout
Ship a change end-to-end (ticket -> spec -> code -> PR -> merge)	Task-delivery pipeline
Look up what a specific /command does and its arguments	Commands reference
Know which skill/persona/agent applies to which work	Skills and agents
Understand the behaviors that apply without being asked	Auto-applying rules

The one-paragraph version

You describe a task; Claude turns it into an AISD Jira ticket (/create-task), drafts a spec on a branch (/start-task), then implements it and runs three in-terminal gates -- code diff, tests, AI code review -- before opening a PR to the repo's integration branch (/approve-spec). A human approves the PR on GitHub; /approve-pr verifies that approval, merges, and advances the ticket. Ten merged tasks roll up into a release PR to master (/release). Throughout, Claude picks the cheapest capable model per phase, consults a token-saving context cache, and reads a deploy registry instead of re-hunting AWS.

Terminal-only flow

The active pipeline runs entirely in the Claude Code terminal -- every approval gate is an in-chat prompt. The older render-UI "bridge mode" is dormant (see Auto-applying rules).