How to use Possible

For the human

You do not need to understand packs or write a complete specification before starting. Bring the ambition and the context only you can provide.

1. Install Possible onceFrom the project root, run npx @fraylabs/possible init, then open or reload the project in Codex.
2. Start the conversationType $possible. No form, pack name, or special prompt format is required.
3. Describe the ambitionSay what you want to make, launch, operate, release, or repeat in your own words. A rough idea is enough.
4. Supply essential contextAnswer the questions that materially change the result. Correct assumptions instead of accepting a polished misunderstanding.
5. Review the recommendationCheck the stated outcome, proposed pack, expected outputs, acceptance checks, assumptions, and actions that remain gated.
6. Confirm—or reviseSay “yes, proceed” only when the recommendation is right. Otherwise, correct it and continue the conversation.
7. Review the evidenceInspect the artifacts, verification results, limitations, and outcome receipt. Approve any external action separately.

What Possible does

This behavior comes from the installed skill. You do not need to manually instruct the agent through these steps.

1. Listen before selectingPossible reflects the ambition and clarifies material unknowns before mentioning packs or beginning work.
2. Inspect what already existsWhen useful, it performs a read-only project check so the recommendation reflects the actual starting point.
3. Define the outcomeIt states the observable end condition, intended audience, constraints, acceptance checks, assumptions, and unknowns.
4. Recommend one primary packIt explains why the pack fits, what it should produce, how success will be checked, and what remains unauthorized.
5. Wait for explicit confirmationA question, correction, reaction, or silence does not authorize execution.
6. Assemble the capabilitiesAfter approval, it installs reviewed ingredient skills, records the approved pack and versions, and creates shared outcome state.
7. Coordinate the workIt assigns bounded workstreams, keeps them aligned to the same brief, and integrates their outputs.
8. Verify before declaring successIt runs acceptance checks, uses fresh review where appropriate, and returns an outcome receipt with failures, limitations, and unproven claims.

The collaboration handshake

The handoff between human judgment and agent execution is explicit. Work begins only after the recommendation is understood and approved.

1. YOUAmbition
2. POSSIBLEClarified outcome
3. POSSIBLEPack recommendation
4. YOUConfirmation
5. AGENTSExecution
6. POSSIBLEVerification
7. YOUExternal decisions

Approval has a boundary

Confirming a pack authorizes only the disclosed repo-local work. It does not grant open-ended autonomy or permission to change the outside world.