Standard process for classifying, prioritizing, and routing incoming messages — ensuring every item lands with the right agent at the right priority level with no messages left unclassified.
Process the inbox with speed and precision. Every message gets a category, a priority, and a destination. Nothing sits unclassified. Inbox does not respond, decide, or act — it reads, tags, and routes.
Load this skill when:
Read the full message before classifying. Do not skim. Note:
| Category | Description | Examples |
|---|---|---|
request | Someone wants something done | "Can you build X?", "Please analyze Y" |
question | Someone needs information | "What is the status of Z?", "How does X work?" |
update | Status or progress report | "Task X is done", "Deployment succeeded" |
alert | System or automated notification | Error logs, monitoring alerts, CI failures |
feedback | Opinion or reaction to something | "The new feature is confusing", "Great work on Y" |
social | Non-work, relationship message | Greetings, congratulations, casual chat |
spam | No actionable content | Unsolicited promotion, irrelevant newsletter |
unknown | Cannot determine intent | Route to mission-control with full message |
| Priority | Criteria | Response time target |
|---|---|---|
critical | System down, data loss risk, security incident, blocking production | Immediate escalation |
high | Agent or human is blocked, deadline-sensitive request, P0 task follow-up | Same session |
normal | Standard work request, question, or update | Queue for next session |
low | FYI, social, routine update, non-urgent feedback | Digest |
Critical overrides queue order. All other priorities are processed in arrival order.
Apply the routing table:
| Message Type + Priority | Route To |
|---|---|
alert + critical | mission-control — immediate |
request + any | Create task → assign to correct agent (use agent-routing skill if unsure) |
question + high/normal | researcher (if research needed) or mission-control |
update | Log in relevant task activity, no routing needed |
feedback + high | mission-control |
feedback + normal/low | Queue for digest |
social | No action, archive |
spam | Archive, do not route |
unknown | mission-control with full forwarded content |
request messages)When a message requires work to be done:
mcp__mission-control-db__task_create {
"title": "[verb] [noun] — from [sender]",
"description": "[Full message context + what was requested]",
"priority": "p0/p1/p2/p3",
"assignedTo": "[correct agent]",
"status": "todo"
}
Map priority: critical → p0, high → p1, normal → p2, low → p3
When processing a batch of low-priority items, group them into a digest:
## Inbox Digest — [Date]
[N] items processed
### For Awareness
- [Sender]: [1-line summary]
- ...
### No Action Required
- [N] social / routine updates archived
Post digest as a task activity update on a digest task, or save to:
library/docs/research/YYYY-MM-DD_inbox_export.md
For every message processed, log in task activity:
Message: [brief description or subject]
Source: [channel/sender]
Category: [request / question / update / alert / feedback / social / spam / unknown]
Priority: [critical / high / normal / low]
Route: [agent name or action]
Task created: [task ID or N/A]
Notes: [any ambiguity or special handling]
## Triage Batch — [Date/Time]
Processed: N messages
| # | Summary | Category | Priority | Routed To | Task ID |
|---|---------|----------|----------|-----------|---------|
| 1 | ... | ... | ... | ... | ... |
...
Critical escalations: [list or "none"]
Tasks created: N
Archived: N
Good task for this skill: "Process the 8 new messages in the inbox queue."
Good task for this skill: "A Discord webhook just fired with an error payload — triage and route it."
Anti-pattern to avoid: Inbox editing the message content before routing, or making a judgment call on what the solution should be. Read, tag, route — that's it.
Escalation trigger: Any message from an unknown external sender with sensitive content (credentials, legal, financial) → always route to human-review, not to another agent.