菜单
Classify incoming emails into triage categories using content analysis, sender profile memory, and thread detection. Produces a category (URGENT, REPLY_NEEDED, FYI, NEWSLETTER, NOTIFICATION, SPAM) with confidence and reasoning.
Translate natural language image descriptions into detailed, structured DALL-E prompts with subject, style, composition, lighting, and mood specifications.
Decompose mathematical problems into sub-expressions, evaluate each one with the calculator tool, and present the full working chain. Handles arithmetic, trigonometry, logarithms, and financial formulas.
Compare two or more PDF documents by extracting targeted sections, building a structured comparison matrix, and highlighting differences with page references.
Extract structured data from web pages using browser snapshot and text tools, then process it into tables, comparisons, or summaries using Python.
Analyze endpoint latency trends using historical check data from memory. Detects slow degradation, spikes vs sustained issues, and calculates baseline deviations.
Validate API response structure and content. Detects schema drift, unexpected null values, and abnormal response sizes.
| name | email-triage |
| description | Classify incoming emails into triage categories using content analysis, sender profile memory, and thread detection. Produces a category (URGENT, REPLY_NEEDED, FYI, NEWSLETTER, NOTIFICATION, SPAM) with confidence and reasoning. |
Email triage classification skill.
Use this skill when processing a new email that needs to be classified for priority and action required.
| Category | Action | Slack alert |
|---|---|---|
| URGENT | Needs immediate attention | Individual, immediate |
| REPLY_NEEDED | Requires a response | Individual |
| FYI | Informational, no action | Batch digest (3+) |
| NEWSLETTER | Subscription content | Batch digest (3+) |
| NOTIFICATION | Automated system notice | Batch digest (3+) |
| SPAM | Unwanted, unsolicited | Never |
Read the full message with read_email. Note:
Recall the sender from semantic memory:
recall("<sender email> sender_profile")
If a profile exists, use it to inform classification:
If no profile exists, create one after classification:
remember("<name>, <email>, first seen <date>, context: <topic>",
category="sender_profile")
Scan the email body for signal indicators:
URGENT signals:
REPLY_NEEDED signals:
FYI signals:
NEWSLETTER signals:
NOTIFICATION signals:
SPAM signals:
Check if the email is part of an existing thread:
Thread context affects classification:
Recall custom triage rules:
recall("<sender> triage_rule")
recall("<subject keyword> triage_rule")
User-defined rules always override the default classification.
Use the think tool to reason through the classification:
If confidence is low, default to FYI rather than over-alerting.