teams-monitor

name	teams-monitor
description	Monitor Microsoft Teams channels via WorkIQ for actionable messages directed at the squad, and bridge them into GitHub issues
domain	communication-bridge
confidence	low
source	manual

Context

The squad has two-way Teams communication:

• READ: WorkIQ (workiq-ask_work_iq) reads Teams messages
• SEND: Teams Incoming Webhook sends messages to Teams

The webhook URL is stored as an environment variable TEAMS_WEBHOOK_URL or can be invoked directly via Invoke-RestMethod.

Sending Messages to Teams

The webhook URL is stored at C:\Users\tamirdresher\.squad\teams-webhook.url (NOT in the repo — it's a secret).

To send a message to Teams, read the URL from that file and POST:

$webhookUrl = Get-Content "$env:USERPROFILE\.squad\teams-webhook.url" -Raw
$body = @{ text = "Your message here" } | ConvertTo-Json
Invoke-RestMethod -Uri $webhookUrl.Trim() -Method Post -ContentType "application/json" -Body $body

IMPORTANT: Never hardcode the webhook URL. Always read from ~\.squad\teams-webhook.url.

Trigger

• Scheduled (Primary): Runs every 20 minutes via Ralph's loop using .squad/schedule.json and Squad Scheduler (task ID: teams-message-monitor)
• Session start: Any agent session can run the Teams check as part of initialization
• On-demand: Any agent can invoke this skill when asked to "check Teams"

The scheduled check uses the script at .squad/scripts/teams-monitor-check.ps1 which is dispatched by the Squad Scheduler engine.

Workflow

Step 1: Query WorkIQ for Recent Messages

Run targeted WorkIQ queries to find messages relevant to the squad. Use the workiq-ask_work_iq tool with questions like:

"What messages were sent in the last 24 hours in Teams channels related to DK8S, idk8s, or Defender Kubernetes?"

"Are there any recent Teams messages from Tamir Dresher or the DK8S team mentioning squad, AI agents, research, or action items?"

"What are the most recent Teams messages in channels I'm part of that mention infrastructure analysis, Aurora, ConfigGen, or fleet management?"

Adapt queries to the squad's active work. Check decisions.md and open GitHub issues for current topics, then query for those topics specifically.

Step 2: Filter for Actionable Content

Not every Teams message is actionable. Look for:

• Direct requests: "Can you look into...", "We need...", "Please investigate..."
• Questions awaiting answers: "Has anyone figured out...", "What's the status of..."
• Decisions or announcements that affect squad work: policy changes, priority shifts, new requirements
• Mentions of squad-related keywords: squad, agent, Picard, research, analysis
• Escalations or incidents: Sev2, outage, breaking change, urgent

Ignore:

• Social/casual messages
• Messages the squad has already processed (check existing GitHub issues for duplicates)
• Automated notifications (build results, PR notifications — these are already in GitHub/ADO)

Step 3: Create GitHub Issues

For each actionable item, create a GitHub issue with:

• Title: [Teams Bridge] <concise summary of the request>
• Labels: teams-bridge, plus any relevant domain labels
• Body must include:
  • Source: who said it, approximate when, which channel/thread (as reported by WorkIQ)
  • The actual request or question, quoted
  • Any context or urgency indicators
  • ---
  • 🔗 Bridged from Teams via WorkIQ by teams-monitor skill
  • 📅 Detected: <timestamp>

Step 4: Deduplicate

Before creating an issue, search existing GitHub issues for:

• Similar title text
• The teams-bridge label
• Recent issues (last 7 days) covering the same topic

If a match is found, add a comment to the existing issue instead of creating a duplicate.

Step 5: Log Activity

After each check, log what was found (or that nothing was found) so the team knows monitoring is active. This can be a brief note in the orchestration log or session output.

Queries Reference

These are starting-point queries. Agents should adapt based on current squad context:

Purpose	WorkIQ Query
General channel check	"What recent Teams messages mention DK8S, squad, or AI agents?"
Tamir's directives	"What did Tamir Dresher say recently in Teams about priorities or action items?"
Incident awareness	"Are there any recent Teams messages about outages, Sev2 incidents, or urgent issues related to Kubernetes or DK8S?"
Meeting follow-ups	"What action items came out of recent DK8S team meetings?"
Topic-specific	"Any recent Teams discussion about [current topic from decisions.md]?"

Limitations

• Read-only: WorkIQ cannot post to Teams. Responses must go through GitHub issues/comments
• Query freshness: WorkIQ results may have indexing delay (typically minutes, occasionally hours)
• No real-time streaming: This is poll-based, not event-driven. Messages may be picked up with delay
• Channel visibility: WorkIQ can only see channels the authenticated user has access to
• No thread-level precision: WorkIQ may not distinguish individual thread replies — treat results as approximate
• First implementation: Confidence is LOW. Expect iteration on query patterns, filtering heuristics, and deduplication logic

Anti-Patterns

• Spamming WorkIQ: Don't run queries more than once per session or per Ralph cycle. Rate-limit yourself
• Creating issues for everything: Be selective. Not every Teams message is an action item
• Ignoring duplicates: Always check for existing issues before creating new ones
• Hardcoding channel names: Channels change. Use topic-based queries, not channel-name queries
• Treating WorkIQ as real-time: It's indexed search, not a live feed. Design for eventual consistency

Future Evolution

1. Confidence → medium: After 2-4 weeks of operation, tune queries and filters based on false positive/negative rates
2. Bidirectional bridge: When Teams bot or webhook becomes available, extend to push GitHub activity summaries to Teams
3. Smart prioritization: Use message sentiment, sender role, and keyword urgency to auto-prioritize bridged issues
4. Channel-specific profiles: Build per-channel query templates as the team learns which channels matter most