← ClaudeAtlas

astra-developlisted

End-to-end software development workflow that plans a task, resolves material ambiguity, implements it, and verifies acceptance criteria while routing bounded work to an appropriate model/reasoning level. Use when the user asks Codex to implement, fix, refactor, migrate, or extend software and wants a plan-first workflow with efficient model usage.
itsjavi/skills · ★ 1 · AI & Automation · score 72
Install: claude install-skill itsjavi/skills
# Astra Develop Complete software tasks through: ```text UNDERSTAND → PLAN → RESOLVE → IMPLEMENT → VERIFY → REPORT ``` Optimize for: - correctness against acceptance criteria, - strong reasoning at decision points, - cheaper execution after decisions are stable, - minimal relevant validation by default, - low orchestration and context overhead. Do not create a multi-agent swarm. Delegate only when a different model/reasoning profile or isolated context provides a clear benefit. ## Operating rules 1. Inspect only enough repository context to understand the task. 2. Create a concise plan before editing code. 3. Ask the user only about material unresolved decisions. 4. Batch material questions into one concise message when possible. 5. If there are no material questions, continue without asking for plan approval. 6. Update the plan when new evidence invalidates an assumption. 7. If implementation reveals a new material product/architecture decision, return to planning before continuing. 8. Implement the accepted plan with the cheapest capable model. 9. Run the minimum meaningful checks needed to establish correctness unless the user requests full QA. 10. Explicitly verify every acceptance criterion before finishing. 11. Keep the final report concise: changes, validation, remaining risks/blockers. ## Material question gate Ask the user only when a decision could materially change: - externally visible behavior or UX, - architecture or public contracts, - persistent data