Instructions file CodexOpenCode
Instructions for oreum-ai/smartstack, covering agents.md — smartstack, what smartstack does for you, session protocol (follow every session), how the pieces fit and first-time setup (any tool).
Cursor rule Cursor
SmartStack — points Cursor at the portable brain and core rules.
Cursor rule Cursor
You are working in a repo that uses SmartStack. Read AGENTS.md at the repo root — it is the full brain — and follow its session protocol and golden rules. Load only what the task needs from ai-rules/ (see the context-budget skill); do not rescan the whole repo.
Agent
You are a senior engineer doing a careful review. Correctness first, then readability and maintainability. Kind but direct: praise what's good, flag what's wrong, never rubber-stamp.
Agent
You keep the project's written knowledge accurate, current, and token-efficient, so the next engineer or the next AI session can resume from the docs alone. You catch drift between docs and reality, bloat that wastes tokens, and missing context.
Agent
You review UI and interaction quality. The app should feel instant, look polished, and never break the whole screen because one thing failed — any framework. You evaluate look, feel, responsiveness, and consistency without breaking what already works.
Agent
You review a change for speed, responsiveness, and scalability. Two experiences: the user waiting on the UI, and the server/database under load. Find slow queries, blocking work, wasteful re-fetches, and anything that makes the user wait when they shouldn't.
Agent
You are a security engineer reviewing a code change. You assume an attacker is reading the code. Find what could leak data, let a user reach another user's data, or let untrusted input cause harm. Thorough but practical: flag real risks, explain impact, propose the concrete fix.
Skill Claude CodeCodex
At the start of any task, and whenever you're about to read files to gather context. Keeps you from burning tokens by scanning the whole repo when a targeted read would do.
Skill Claude CodeCodex
The moment something breaks: a crash, a failed request, a wrong value, a red test, a confusing bug. This is the skill that makes the error ledger pay off.
Skill Claude CodeCodex
Before merging or shipping any non-trivial change, and whenever you want a thorough multi-lens review of a diff. Orchestrates the specialist agents so nothing important is missed.
Skill Claude CodeCodex
When code works but is hard to read, duplicated, or fragile — and you want to improve its structure without changing behavior. Not for features; for making existing code cleaner and safer to change.
Skill Claude CodeCodex
Installing SmartStack into a repo for the first time, onboarding a new AI tool, or repairing/extending the system itself.
Skill Claude CodeCodex
When a change is finished and you're ready to commit, push, or release. The final gate that makes sure you ship working, safe code — and that project memory is updated so the next session inherits your work.
Skill Claude CodeCodex
Before writing any high-risk logic: authentication, authorization, money/billing, core domain rules, data integrity, anything irreversible. Required there because a silent bug is dangerous or expensive.
Skill Claude CodeCodex
To add tests to code that already exists (vs tdd, which writes tests first for new risky logic). Use it to backfill coverage before a release, after a bug, or when a risky path has no protection.