Where to go from here

You've finished the ten-page handbook. Three directions to take next: ship one, recover when stuck, contribute when ready.

Ship one

The fastest way to internalize the loop is to put a real ticket through it. Pick something small from your queue. Open a Conductor workspace. Type:

> /chowly:implement_jira TEAM-XXXXX

By the third or fourth ticket, the loop becomes muscle memory. Don't skip the small ones early on — they're cheap reps that build the right habits before you take on something bigger.

Recover when things go wrong

Sometimes a session breaks. Claude's looping, the diff is sprawling, the context's full, you've lost the thread. The recovery prompts:

• "List every file you've edited in this session, with a one-line summary of each." Re-establishes ground truth.
• "Pretend this is turn 1. The ticket is TEAM-XXXXX. What would you do first?" Forces a re-frame.
• "Stop. One question only. [narrow factual question]." Re-anchors with a small win.
• "Walk me through the last five turns. What did you do? What did I ask?" Useful when you've lost track yourself.

If those don't work, abandon. Save a session summary to tmp/, open a fresh workspace, read the summary, continue. The cost of abandoning is small. The cost of fighting an overloaded session is large.

Contribute when ready

Once the loop feels natural — usually after a couple weeks — you'll start noticing the same frustrations recurring. "Claude always reaches for webmock when our convention is service-level stubs." "Claude doesn't know about our service object pattern." "Claude needs to be reminded about our migration conventions every time."

Each one is a candidate for either:

• An addition to your repo's CLAUDE.md — the simplest fix, persistent across sessions for everyone.
• A new skill — when the pattern is reusable across multiple repos.
• A new slash command — when the pattern is multi-step and routine.
• A new agent — when the pattern needs a different frame than the existing agents provide.

Senior engineers should be writing skills and commands, not just consuming them. The team's stack improves from inside the team.

Where the full reference lives

This handbook is the curated path. The full library is the reference:

• Workflows — solving a ticket, writing a ticket, debugging, log investigation, perf, refactoring, code review, PR, failed CI.
• Tools reference — Claude Code, Conductor, Atlassian MCP, Context7, Chrome MCP, LSP, jira CLI, gh CLI, Datadog, AppSignal.
• Agents reference — deeper dives on the four work-horses and the built-ins.
• Skills reference — jira, conductor, datadog, appsignal.
• Commands reference — implement_jira, write_specs, fix_specs, cleanup, code_review, request_review, batch_implement.
• Prompting playbook — principles, ticket-research, implementation, corrections, recovery, anti-patterns.
• Case studies — six real tickets walked through, plus per-project overviews.
• Best practices — context management, plan-first, when to spawn agents, conductor workspaces, ticket quality, PR quality, error recovery.
• Reference — glossary, FAQ, troubleshooting.

The way of life

By month three, you'll have four to ten Conductor workspaces in flight at any moment. You won't memorize the codebase, because you don't have to. You'll spend more of your day on judgment — design calls, code review of teammates, conversations with stakeholders — and less on plumbing. The shape of your week will be different.

That's not a goal you reach by reading. It's a place you arrive at by shipping the next ticket, then the next, then the next. The handbook is over. Open Conductor.