j-feature-workflow

name: j-feature-workflow description: "Structured workflow for feature-sized or risky changes: short intake questions, create a branch, use TDD where applicable, commit frequently, get an external agent review, fix issues, then get a second external review and optionally open a PR. Use when the user says a task is big, multi-step, high-risk, or explicitly wants TDD, frequent commits, or multi-agent review."

/j-feature-workflow

Overview

Use this skill to handle large changes safely: a quick intake to clarify ambiguity, then branch isolation, test-driven iterations, frequent commits, and two external reviews before merge or PR.

Intake (Ask First If Ambiguous)

Ask 2–4 short questions if the task has unclear scope or risk.

• What outcome defines “done”?
• Any files/areas to avoid or prioritize?
• Expected tests or constraints?
• Any deadline or review expectations?

Workflow

1. Check docs/TODO.org — if this task is listed, work from that item
2. Create a feature branch named for the task
3. If feasible, start with a failing test (TDD)
4. Implement in small increments and commit often
5. Run pragmatic tests before each review
6. Get a first external agent review (focus on correctness). Ask for the highest reasoning model
7. Apply fixes and commit
8. Get a second external agent review - again, ask for the highest reasoning model
9. Apply fixes, commit, and optionally open a PR
10. Mark the TODO.org item DONE if applicable

Branching

• Create a branch from main (or the user’s target base).
• Keep history linear on the branch; rebase onto base before merge.

TDD Guidance

• Prefer small testable slices.
• If TDD is impractical (e.g., infra or docs), explain why and proceed with targeted checks.

Commit Discipline

• Commit after each coherent slice of work.
• Use clear, descriptive messages (e.g., feat: add sync hooks, fix: handle dangling symlink).

Reviews

• Review 1: External agent, correctness-focused.
• Fix issues, commit.
• Review 2: External agent, correctness-focused.
• Fix issues, commit.

Optional PR

• If the user wants a PR: open a draft PR after the second review is clean.
• Include summary, test results, and any remaining risks.

Notes

• Use the review skill for external agent reviews when available.
• Keep each review prompt terse and focused on bugs/regressions.
• When calling another agent CLI, unset ANTHROPIC_API_KEY and OPENAI_API_KEY so it uses CLI auth (see docs/agent-ops.md "Cross-Agent Reviews").