Getting Started
Choose a path to set up AEP, then run your first save/apply loop.
Pick your setup option
There are two ways to get AEP into your workflow. Choose the one that matches how your team adopts new tools.
Option A (auto): Install the Robi Labs skill, then generate AEP structure
Use this when you want the fastest path with minimal manual file creation.
1) Install
Tell your agent:
Install the Robi-Labs “aep” skill from Smithery for this project, then confirm it’s available.Confirm install details at Smithery skill page.
Or install directly for a specific agent using the Smithery CLI:
2) Prepare your project AEP structure
Tell your agent to inspect the repo and create the minimal AEP files in agent-visible paths:
Inspect this repository. Create a project-level AEP pack (and any required supporting files) so other tasks can “save/apply” immediately.
Use the AEP spec from /docs/file-spec.
Prefer agent-specific directories (e.g. .claude/, .codex/, .cursor/) when appropriate.3) Run your first save/apply loop
First, complete a small task successfully and capture what worked:
Save this successful task as a task-level AEP pack with:
- intent
- constraints
- workflow
- failure_traps (from what went wrong or what you avoided)
- success_checks (objective, verifiable)Then apply it on the next similar task:
Before implementation, apply the most relevant AEP pack for this task.
For each workflow step, explain what you’re doing and how you’ll validate success checks.
After completion, confirm which success checks passed and propose pack improvements.Option B (manual): Create AEP files yourself (or clone the starter pattern)
Use this when you want full control over what’s added to your repo, or when your environment can’t install skills automatically.
1) Create the project pack file
Tell your agent (or do it yourself) to create your pack at a path that’s visible to the active agent.
Example starting point:
- create
.aep/project.aep.json
Use the structure from AEP File Structure / Spec.
Minimal template:
{
"intent": "Deliver scoped changes safely.",
"workflow": ["Inspect", "Implement", "Verify", "Report"],
"success_checks": ["Requested behavior works", "No new type errors"]
}2) (Optional) Add/Update agent instruction files
Tell your agent to ensure your instruction files explicitly reference AEP usage:
Inspect this repo for agent instruction files (e.g. AGENTS.md, CLAUDE.md, and any agent-specific instruction files).
Update them so agents prioritize applying AEP before starting work, and so they save successful outcomes back into AEP packs.3) Run your first save/apply loop
Save a successful task:
Save this successful task as a task-level AEP pack. Make success checks objective and easy to validate.
If anything went wrong, convert it into failure_traps.Apply the pack on the next task:
Apply the most relevant AEP pack before implementation and follow its workflow steps.
After finishing, report each success check result and propose updates to the pack.Verify it’s working
After your first apply:
- the agent loaded the pack before implementing
- the workflow steps were followed in order
- success checks were executed and reported
- the output matches the intent (not just “it seems fine”)
Last updated on