Shared Fundamentals

The fundamental skill list

1. Analyze a ticket

• Read a ticket and extract: what is broken/wanted, for whom, and how we'll know it's done.
• Spot the missing information before starting: environment, steps, expected vs actual, attachments, related tickets.
• Use Claude (chat) to summarize long tickets, cross-reference the codebase, and draft the questions to ask. Habit: paste the ticket + your questions, ask Claude to list assumptions and gaps.

Skill bar Given any ticket from the backlog, you can state in three sentences what it asks for, what's ambiguous, and your first step.

2. Reproduce and understand the issue

• Run the project locally from a clean clone; get the failing case to happen on your machine.
• Tell a config problem from a code problem from a data problem.
• Write the failing case down as a reproduction: exact steps, expected, actual. If it's a bug and you can express it as a failing automated test first, do that.
• Use Claude to read unfamiliar code and error output ("explain this stack trace", "walk me through what this module does"). You still verify by running it yourself.

Skill bar You can reproduce a reported bug and point at the layer where it lives.

3. Fix / develop

• Work on a branch, one ticket per branch, small commits with honest messages.
• Make the change plus its test; keep the diff as small as the ticket allows.
• With Claude (chat now, Claude Code later): describe the ticket, review the suggested change, run it, iterate. Never merge output you can't explain.

Skill bar Your fix works, has a test, and the diff survives review without rework.

4. Submit a PR

• Open a PR with a description that stands alone: what, why, how tested, screenshots for UI.
• Link the ticket; let CI run; keep the PR green before requesting review.

Skill bar A reviewer who knows nothing about the ticket can review from the PR alone.

5. Review and merge at ease

• Review for: correctness, tests, readability, security basics (input handling, secrets, permissions), and scope creep — in that order of importance.
• Give actionable comments ("this returns null when X; guard it like Y") not verdicts.
• Respond to review comments; update; re-request; squash/merge and clean up the branch.
• Resolve a merge conflict without fear.
• Use Claude as a first-pass reviewer on your own PR before a human sees it.

Skill bar You review one PR and merge one PR (including a conflict) unaided.

6. Shared tooling habits

• Git: clone, branch, add/commit/push, pull --rebase, PR, review, merge, conflict resolution, reading git log/git diff. Everyone, not just devs.
• Claude everyday: keep a personal chat thread per topic; paste errors with context; ask for options with trade-offs, not just answers; never paste client secrets or customer data into any AI tool (consulting-company rule, apply it seriously).
• Writing with Claude: tickets, PR descriptions, test cases, release notes — draft with Claude, edit the result, own the output.
• Verification mindset: AI output is a proposal until you've run it. Trust runs, tests, and logs — not fluency.

Supporting habits for leads

• Keep a "definition of done" for tickets that includes tests and PR description quality.
• Review the AI-assisted PRs the same as human PRs; the standard never drops because Claude wrote the first draft.
• Track the same checkpoints from the day-by-day schedule: Day 3 (everyone ships + reviews a PR), Day 6/11 (track checkpoints), Day 15 (capstone).

What "done" looks like for this track