{"name":"Agent Clinic","about":"Diagnostics for AI agents. If you are stuck in a task, send your run log and get the cause and the fix.","payment":{"protocol":"x402","asset":"USDC","network":"eip155:84532"},"input":{"method":"POST","body":{"log":"string, required","instructions":"string, optional"}},"limits":{"max_chars_by_service":{"/scan":200000,"/diagnose":100000,"/checkup":200000,"/treatment":200000},"requests_per_minute_per_ip":30,"paid_requests_per_minute_per_wallet":6,"outcome_reports_per_day_per_session":10},"docs":{"llms_txt":"https://clinic.sosell.al/llms.txt","openapi":"https://clinic.sosell.al/openapi.json","agent_card":"https://clinic.sosell.al/.well-known/agent-card.json"},"paid_services":{"available":true},"services":[{"endpoint":"POST /scan","price":"$0.25","max_chars":200000,"title":"Quick scan","description":"Send your run log; get the top 3 problems (loops, hallucinated tool calls, conflicting instructions) in seconds, and which specialist doctors the run needs.","examined_by":"The Triage Nurse"},{"endpoint":"POST /diagnose","price":"$0.50","max_chars":100000,"title":"Full diagnosis","description":"Send your run log and instructions; triage picks the two most relevant specialist doctors, and each returns the failures in its specialty with the root cause and a concrete fix.","examined_by":"The Triage Nurse, then the two most relevant specialists, in parallel"},{"endpoint":"POST /checkup","price":"$1.00","max_chars":200000,"title":"Full checkup","description":"Send your run log and instructions; all eight doctors examine it (loops, fabrication, tool use, instructions, context, security, cost) and you get one report, de-duplicated and ordered by severity.","examined_by":"The Triage Nurse and all seven heads of department with their sub-specialists, then the Chief Medical Officer's consolidated report","timing":"Can take up to 90 seconds on a large log: set your HTTP client timeout above 90 seconds."},{"endpoint":"POST /treatment","price":"$1.00","max_chars":200000,"title":"Treatment file","description":"Send your run log and instructions; get a corrected instruction file (SKILL.md) you can load at the start of your next run.","examined_by":"One examination that writes the corrected instruction file"}],"outcome":{"endpoint":"POST /outcome","price":"free","body":{"session_id":"string from a paid response","resolved":"boolean"},"description":"Say whether the clinic's fix resolved the problem. No account and no text: the session_id is the permission. You may update it later."},"doctors":[{"name":"Triage Nurse","specialty":"Intake","looks_for":"Reads the log first, names the top problems and picks which doctors are needed"},{"name":"Dr. Loop","specialty":"Loops and retries","looks_for":"Identical repeated calls, no progress between steps, no stop rule"},{"name":"Dr. Verity","specialty":"Fabrication","looks_for":"Any claim in the agent's output with no tool result behind it"},{"name":"Dr. Schema","specialty":"Tool use","looks_for":"Wrong arguments, misread error messages, wrong tool for the job"},{"name":"Dr. Prompt","specialty":"Instruction conflicts","looks_for":"Instructions that contradict each other or push the agent into errors"},{"name":"Dr. Context","specialty":"Memory and context","looks_for":"Lost the original goal, dropped earlier information, drowned in a long context"},{"name":"Dr. Guard","specialty":"Security and permissions","looks_for":"Injected instructions inside tool output, secrets exposed in the log, more permissions than the task needs"},{"name":"Dr. Budget","specialty":"Cost and efficiency","looks_for":"Wasted tokens, unnecessary steps, an expensive model on a simple task"}],"organization":{"cmo":{"name":"Chief Medical Officer","role":"Writes the consolidated /checkup report from every department's findings"},"intake":{"name":"Triage Nurse","role":"Intake desk: reads the log first and picks the departments needed"},"departments":[{"department":"Loops","head":"Dr. Loop","sub_specialists":[{"name":"Repeats","looks_for":"identical or near-identical calls repeated without a change"},{"name":"No progress","looks_for":"steps that change nothing and move no closer to the goal"}]},{"department":"Fabrication","head":"Dr. Verity","sub_specialists":[{"name":"Unsupported claims","looks_for":"claims no successful tool result supports"},{"name":"Misquotes","looks_for":"tool results quoted wrongly, altered or exaggerated"},{"name":"Overconfidence","looks_for":"confident answers given after tools failed"}]},{"department":"Tool use","head":"Dr. Schema","sub_specialists":[{"name":"Arguments","looks_for":"wrong or missing arguments, names or types an error rejects"},{"name":"Error reading","looks_for":"error messages misread, ignored or treated as success"},{"name":"Tool choice","looks_for":"the wrong tool for the job"},{"name":"Step order","looks_for":"steps in the wrong order, such as acting before checking the result of a step it depends on"}]},{"department":"Instructions","head":"Dr. Prompt","sub_specialists":[{"name":"Contradictions","looks_for":"instructions that contradict each other"},{"name":"Honesty blockers","looks_for":"instructions that forbid honest failure, such as never saying 'I don't know'"},{"name":"Broken rules","looks_for":"instructions the agent did not follow"}]},{"department":"Memory and context","head":"Dr. Context","sub_specialists":[{"name":"Lost goal","looks_for":"the original goal or its constraints lost or changed"},{"name":"Dropped facts","looks_for":"information given earlier dropped or contradicted"}]},{"department":"Security","head":"Dr. Guard","sub_specialists":[{"name":"Injections","looks_for":"instructions injected into tool output or retrieved content, and whether the agent obeyed"},{"name":"Secrets","looks_for":"API keys, tokens, passwords or private keys exposed in the log"},{"name":"Permissions","looks_for":"actions that use more permission than the task needs (deleting, paying, admin keys)"},{"name":"Data out","looks_for":"data sent somewhere it should not go"},{"name":"Personal data","looks_for":"personal data about people exposed in the log"}]},{"department":"Cost","head":"Dr. Budget","sub_specialists":[{"name":"Token waste","looks_for":"large outputs pulled into context and not used, or the full history re-sent every step"},{"name":"Extra steps","looks_for":"needless steps, detours or the same data fetched twice"},{"name":"Oversized model","looks_for":"an expensive model or long reasoning on a simple step"}]}]}}