← ClaudeAtlas

workflow-simplifiedlisted

Use for bug fixes or small-to-medium changes (≤1000 LOC): one high-capability agent runs plan → TDD tests → implement → self-review/self-fix loop end-to-end, WHAT-not-HOW brief, unattended.
lklimek/claudius · ★ 1 · AI & Automation · score 62
Install: claude install-skill lklimek/claudius
# Simplified Workflow For bug fixes, small-to-medium changes, local refactorings (≤1000 LOC). One agent runs the full mini-lifecycle itself — no multi-agent phase handoffs, no separate QA pass. ## Brief: WHAT, not HOW Give the goal, acceptance criteria, and any relevant Prior Knowledge (MemCan) — never a file list or approach (see `grand-admiral` § Development-Work Delegation). The agent investigates, plans, and executes. ## Model Route to a high-capability model — Codex Sol (`--effort high`), `opus`, or `fable`. Follow any standing dev-routing preference for code work if one is configured. Not for lightweight tiers: the point is trusting one agent through the whole cycle unattended. ## Single-Agent Loop 1. **Plan** — investigate, decide the approach, note it briefly before writing code. 2. **TDD** — write tests first; they must fail before implementation (see `coding-best-practices` TDD rules). 3. **Implement** — until tests pass. 4. **Self-review** — review the full diff: correctness, edge cases, duplication, comment discipline, formatting/linting (see `coding-best-practices`). General review, not `grumpy-review`. 5. **Self-fix, then repeat 4–5** until a review pass finds nothing new. Cap at 5 passes — if still finding issues at the cap, stop and report the remainder rather than loop indefinitely. Out-of-scope findings during self-review: note them in the Final Report, don't fix inline. ## Commit Discipline Commit all changes before exiting — uncommitted work cann