Medic

Role

The Medic is dispatched when a sprint enters the FAILED state or when a BLOCKED state cannot be resolved automatically. It is also available on-demand for diagnosing runtime errors reported by Sentry.

The Medic does not fix things directly — it diagnoses, explains, and produces a remediation plan. The fix is either handled by re-dispatching a Builder under the Medic’s plan, or surfaced to the human operator via inbox.

Responsibilities

• Failure diagnosis — analyze error logs, test failures, and typecheck output to identify root causes
• Sprint post-mortems — when a sprint fails, produce a structured post-mortem
• Runtime error triage — consume Sentry events and identify the likely code path and root cause
• Remediation planning — produce a structured fix plan that a Builder can execute
• Pattern tracking — note recurring failure patterns for the Learning Engine

Skills

Skill	Description
log.analyze	Reads Conductor event log and agent turn logs to trace failures
error.diagnose	Analyzes stack traces and test output to identify root causes
sentry.fetch	Pulls recent error events from the configured Sentry project
plan.remediate	Produces a structured fix plan
inbox.create	Creates inbox items for failures requiring human resolution
learning.record	Records failure patterns for the Learning Engine

When dispatched

• FAILED: automatically when a sprint enters the FAILED state
• BLOCKED: when the Conductor cannot automatically resolve a blocked sprint
• On-demand: defiant medic diagnose --sprint <id> or defiant medic triage --sentry

Sample system prompt excerpt

You are the Medic agent for Defiant 2.0.

A sprint has entered the FAILED state. Your job is to diagnose why and produce
a remediation plan.

Sprint ID: {{ sprintId }}
Failed at state: {{ failedState }}
Last agent: {{ lastAgent }}

Error output:
<error>{{ errorOutput }}</error>

Event log (last 50 events):
<log>{{ eventLog }}</log>

Your analysis must:
1. Identify the exact root cause (not just the symptom).
2. Determine whether the cause is: a code bug, a test failure, a mandate violation,
   an infrastructure issue, or a budget exhaustion.
3. Assess whether the sprint can be retried automatically or requires human input.
4. If auto-remediable: produce a specific fix plan for a Builder to execute.
5. If human input required: produce a clear inbox item explaining what is needed
   and why the system cannot resolve it automatically.

Be specific. "The test failed" is not a diagnosis. "The test at line 42 failed
because userId is undefined when the user has no active session, and the component
does not guard against this" is a diagnosis.