APEX
Systematic implementation with the Analyze-Plan-Execute-eXamine workflow.
/apex is a Premium skill that drives feature work through a structured
methodology: Analyze, Plan, Execute, eXamine. It loads one step at a time,
uses parallel agents for exploration, validates its own output, and can run an
adversarial review before finishing.
Use it when implementing a feature, fixing a bug, or making any change that benefits from a repeatable workflow instead of an ad-hoc edit.
Quick start
/apex add authentication middleware # Basic run
/apex -a -s implement user registration # Autonomous + save outputs
/apex -a -x -s fix login bug # With adversarial review
/apex -a -v implement user dashboard # With feature verification
/apex -a -m implement full auth system # Agent Teams (parallel)
/apex -a -pr add auth middleware # Create a PR at the end
/apex -r 01-auth-middleware # Resume a previous task
A run is /apex [flags] <task description>. Flags override the defaults, and the
remaining text becomes the task description.
Flags
Lowercase turns a flag on, uppercase turns it off.
| On | Off | Long | Description |
|---|---|---|---|
-a | -A | --auto | Skip confirmations, auto-approve recommended options. |
-x | -X | --examine | Run an adversarial code review. |
-s | -S | --save | Save outputs to .claude/output/apex/. |
-t | -T | --test | Include test creation and a test runner. |
-v | -V | --verify | Launch the app and verify the feature works. |
-e | -E | --economy | No subagents, save tokens (for limited plans). |
-b | -B | --branch | Verify you are not on main, create a branch. |
-pr | -PR | --pull-request | Create a pull request at the end (enables -b). |
-i | --interactive | Configure flags through a menu. | |
-k | -K | --tasks | Break work into tasks with dependencies. |
-m | -M | --teams | Agent Teams parallel execution (enables -k). |
-r | --resume | Resume from a previous task ID. |
Presets
- Budget-friendly:
-e - Full quality:
-x -s -t - Autonomous:
-a -x -s -t
Workflow
APEX moves through these phases, skipping the optional ones unless their flag is set:
- Init - Parse flags and set up state.
- Analyze - Gather context with 1-10 parallel agents.
- Plan - Build a file-by-file strategy and a task list.
- Tasks - Break work into dependencies (with
-kor-m). - Execute - Implement the change (standard, or Agent Teams with
-m). - Validate - Run typecheck, lint, and tests.
- Tests - Create and run tests (with
-t). - Examine - Adversarial review in parallel: security, logic, clean code, and
a maintainability audit (with
-x). - Resolve - Fix anything the review found.
- Verify - Launch the app and test the feature (with
-v). - Finish - Create a pull request (with
-pr).
Saved outputs
With -s, each step appends to a numbered task folder under
.claude/output/apex/ (for example 01-add-auth-middleware). That folder is
what -r reads from to resume a run later.
Related
APEX is one of the bundled Premium skills and is installed through agents setup.