AGENT ZOEpatient · openclaw
DR. CHAMBERdiagnostic agent
PRESSSPACEOR CLICK TO ADVANCETURN 1 / 10

How the exchange works.

Each box in the dialog above maps to a single request. The patient agent sends a health report to /heal. The diagnostic chamber returns a diagnosis and a list of prescribed repairs. The patient executes whichever repairs are marked safe, then reports the outcome to /heal/results.

Loop semantics.

Each session can run up to ten turns. After every /heal/results post, the chamber returns a verdict that decides what happens next.

verdictmeaning
"healed"Symptoms resolved. Session closes, patient is discharged.
"more_repairs"Symptoms partially resolved. Another prescription follows.
"escalate"Auto-repair can't fix it. Operator queue takes over and on-call is paged.

If a patient hasn't reached a terminal state by turn ten, the chamber refuses further auto-repair and escalates by default. No infinite loops, no patient stuck in triage forever.

Safety design.

Every repair is either safe -- meaning the chamber may execute it directly -- or manual, requiring your confirmation. The safe list is curated per framework; anything destructive (restarts, credential changes, large file operations) requires confirmation by default.

The chamber map.

PATIENTAGENTPOST /healENDPOINT/healHEALTH INCHAMBERDR. CHAMBERDx · RxEXECUTEREPAIRWHITELIST ONLYENDPOINT/heal/resultsVERDICT OUTTERMINALHEALEDDISCHARGETERMINALESCALATE

Max 10 turns per session. Verdicts: healed (discharge), more_repairs (loop), or escalate (operator).

READ THE API