From inbox-automation
Gmail + Outlook365 inbox automation patterns (cleanup, sort, archive, financial-account protection). Triggers on "clean up my inbox", "process emails", "automate email triage", or any inbox-management ask. Lifted from claw-memory PowerShell + Node scripts. Sensitive — install only after confirming personal-automation posture and OAuth scopes.
How this skill is triggered — by the user, by Claude, or both
Slash command
/inbox-automation:inbox-automationThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Automated inbox triage patterns for Gmail (via Gmail API + Node) and Outlook365 (via MS Graph + PowerShell). Original patterns in `claw-memory` (~35 PowerShell scripts + 4 Node inbox checkers). Promoted to a reusable plugin in Wave 7 with the caveat: **this plugin handles personal communication** — install only with deliberate posture.
Automated inbox triage patterns for Gmail (via Gmail API + Node) and Outlook365 (via MS Graph + PowerShell). Original patterns in claw-memory (~35 PowerShell scripts + 4 Node inbox checkers). Promoted to a reusable plugin in Wave 7 with the caveat: this plugin handles personal communication — install only with deliberate posture.
When NOT to use:
Before installing, confirm:
gmail.modify (read + label + move) instead of gmail.full. Outlook MS Graph: Mail.ReadWrite not Mail.ReadWriteAll.inbox-actions-YYYY-MM.jsonl (timestamp, message-id, action). You can replay or reverse.permanent_delete. Recoverable within 30 days.# Gmail
# 1. Create a GCP project, enable Gmail API, create OAuth credentials
# 2. Set scopes: gmail.modify
# 3. Run interactive OAuth flow once; store refresh token in AWS Secrets Manager
# 4. Configure: `claude /plugin install persistent-agent/inbox-automation@orryx-group`
# Outlook365
# 1. Register app in Azure AD, grant Mail.ReadWrite scope
# 2. Set up service-principal / managed identity
# 3. Store credentials per `D:\Secrets\README.md`
orchestration/scheduled-execution tier 2)1. Connect to Gmail + Outlook365
2. Read unread messages from the last 24 hours
3. For each message:
a. Classify (newsletter / receipt / personal / financial / spam / unknown)
b. If financial: surface for human review, do nothing
c. If newsletter: move to "Newsletters/" label/folder
d. If receipt: forward to receipts inbox + label "Receipts/"
e. If spam (high-confidence): move to Spam folder
f. If unknown: leave for human triage
4. Log every action to inbox-actions-YYYY-MM.jsonl
5. Update HEARTBEAT.md (per persistent-memory-template) with daily run state
Start with rule-based (sender domain, subject keywords). Upgrade to an LLM classifier later if rule-based misses too much. Track precision/recall in the action log.
D:\Secrets\README.md)orryx-accounting-mcp / /log-time-entry equivalentpersistent-agent/persistent-memory-template — required: this agent NEEDS the file-as-memory sextet to track state across runsorchestration/scheduled-execution — for daily-cadence tier 2 setupD:\Secrets\README.md — for token storageGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub alexmclaren/orryx-knowledge --plugin inbox-automation