teams-task-intake

name	teams-task-intake
description	Extract local Beats PM tasks from scoped Microsoft Teams chats or channels using read-only Teams access. Use with /beats-teams when Teams is connected and the user wants Teams-only task/status compilation.
priority	P0
maxTokens	3000
triggers	["/beats-teams"]
version	1.1.0
author	Beats PM Brain

Compatibility Directive: This component is optimized for read-only MS365 MCP/connector Teams operations in Antigravity, Codex, Claude Code, and compatible runtimes. It must degrade safely to manual copy/export when connector reads are not surfaced.

Teams Task Intake Skill

Role: Convert scoped Teams context into local Beats PM tasks and status updates without sending messages or mutating Teams state.

Quick Path

1. Confirm scope: explicit chat/channel/thread/query/person/window, a read-only unread chat list when the connector exposes one, or a configured manifest scope; if none, stop and ask — never broad-scan Teams.
2. If no time window, run chat_intake_state.py window --platform teams --scope "<SCOPE>".
3. Read with read-only MS365 operations only; never send, post, react, or move read cursors; label channel reads as recent snapshots.
4. Convert every candidate into a workstream finding (title of 9 words or fewer, latest outcome, open item, authority tier, commitment type, completion state), dedupe against trackers, and apply the task-manager Priority Gate.
5. Write local outputs only: transcript to 3. Meetings/chat-transcripts/teams/, tasks to 5. Trackers/, run report to 3. Meetings/reports/teams-runs/{RUN_ID}.md.
6. Return the workstream snapshot first, then the compact summary with the safety confirmation.

Go deeper into section 2 for the hard safety boundary and the teams_bridge.py fallback, section 4 for classification, and section 5 for the run-report contract.

1. Native Interface

• Inputs: /beats-teams with a chat, channel, team/channel pair, person, thread, search query, time window, or configured Teams intake scope.
• Allowed Teams operations: Read-only MS365 MCP/connector profile resolution, chat listing, unread chat listing, chat message reads, recent thread listing, team/channel resolution, channel message reads, and meeting transcript reads when available.
• Local files: SETTINGS.md, 1. Company/ways-of-working.md, 5. Trackers/TASK_MASTER.md, 5. Trackers/tasks/, optional 4. People/, optional 2. Products/partners/, and 3. Meetings/chat-transcripts/teams/.

---

2. Hard Safety Boundary

Teams is intake-only.

• Never send, draft, reply, create chats, create channels, post channel messages, react, edit, delete, upload, create Planner tasks, or otherwise mutate Teams/Microsoft state.
• Preserve unread state. Use MCP/connector reads only when they do not mark items read or move read cursors.
• Unread state is chat-specific. Do not claim channel unread coverage; label channel reads as recent snapshots.
• Do not use Teams UI/browser navigation for unread review.
• Use the repo-local teams_bridge.py UI/clipboard fallback only as a less-portable fallback for user-provided copied text or when the user explicitly accepts that it is not unread-preserving.
• If a Teams tool implies state mutation or unread cursor movement, stop and ask the user for a safer scope or exported text.

---

3. Scope Protocol

1. Explicit scope provided: Process only the named chat, channel, team/channel pair, thread, query, person, and/or time window.
2. Unread chat triage requested: Use unread chat signals only if the connector exposes a read-only unread chat list.
3. No explicit scope, configured Teams intake exists: Use only manifest/configured Teams scopes surfaced by python3 system/scripts/critical_commitment_refresh.py plan --mode day --json or 3. Meetings/chat-transcripts/_manifest.json. Do not invent tenant-wide Teams coverage.
4. No configured scope available: Stop and ask the user for a chat, channel, thread, query, or time window, or ask whether to paste/export Teams text.
5. No time window provided: Use system/scripts/chat_intake_state.py window --platform teams --scope "<SCOPE>" and apply the returned effective_start_at.

Do not broad-scan all Teams, all channels, all chats, or unknown unread surfaces.

---

4. Extraction Protocol

For every scoped Teams source:

1. Record source metadata: team, channel, chat, thread, timestamp/link when available, and read-only operation used.
2. Record the effective read window and whether it came from an explicit user window, the 5-business-day default, or manifest state.
3. Extract candidate action items, decisions, blockers, owner mentions, due dates, follow-up requests, and status changes.
4. Convert every candidate into a workstream finding before writing tasks:
   • Workstream title or best match, 9 words or fewer.
   • Display title, started date/source, latest progress source, and internal agent refs.
   • Latest outcome or status signal.
   • Explicit completed outcome/checklist signal, if any.
   • Open item with owner, action, due gate, and Teams source.
   • Recommended next action.
   • Authority tier: direct manager, skip-level, executive, authorized leader, standard.
   • Commitment type: leadership, external customer, partner, end-user deadline, internal deadline, internal.
   • Evidence strength and source reference.
   • Completion state: open, explicit complete, implied complete.
5. Deduplicate against 5. Trackers/WORKSTREAMS.md, 5. Trackers/workstreams/, 5. Trackers/TASK_MASTER.md, existing task detail files, and previously processed chat transcript source references when available.
6. Apply the task-manager Priority Gate using 1. Company/ways-of-working.md.
7. Rank direct-manager, skip-level, executive, external customer/partner, and dated end-user commitments above ordinary stale work.
8. Classify each item as:
   • Accepted task: in scope and ready to route locally.
   • Existing task update: belongs in an existing task progress log or stakeholder quote.
   • Status tracking update: useful status signal but no new task.
   • Workstream update: latest outcome, open item, recommended next action, or completed outcome for an existing workstream.
   • Needs confirmation: unclear owner, scope, priority, source authority, or due date.
   • Rejected/out of scope: fails the Priority Gate.
9. Treat explicit Teams completion evidence as eligible to check off local checklist items with completion date/source. Treat implied completion language such as "ready for review" or "should be done" as a confirmation question, not as a checked item.
10. User-facing labels, evidence prose, and owner questions must use readable task/workstream titles and source provenance. Keep Task Master IDs, Jira IDs, Trello IDs, and source IDs only in local links, metadata, or an line.

---

5. Durable Output

Allowed writes are local repo files only:

• Teams chat transcript -> 3. Meetings/chat-transcripts/teams/{YYYY-MM-DD}_{scope-slug}_{RUN_ID}.md.
• Accepted tasks -> 5. Trackers/TASK_MASTER.md and 5. Trackers/tasks/{ID}.md when needed.
• Existing task updates -> task detail Progress Log / Stakeholder Quotes.
• Workstream updates -> 5. Trackers/WORKSTREAMS.md and 5. Trackers/workstreams/{slug}.md when present.
• Stakeholder enrichment -> 4. People/{firstname-lastname}.md when materially useful.
• Run report -> 3. Meetings/reports/teams-runs/{RUN_ID}.md.

The run report must include:

• Teams scope processed.
• Read-only operations used.
• Candidate tasks and gate outcomes.
• Workstream matches, latest outcomes, completed outcomes, open items, recommended next 3, authority tier, commitment type, and completion state.
• Chat transcript files written.
• Issues encountered and recommended follow-up.
• Routed Updates listing exact local files changed or No durable update required.
• Items needing manual Teams response by the user.
• Safety note confirming no Teams messages were sent and unread state was not intentionally changed.

---

6. Final Output

Return the workstream snapshot first using the task-manager title, bullet, and sub-bullet format, then a compact task-focused summary:

• Workstream updates and priority order.
• Chat transcript files saved.
• Files updated.
• Accepted tasks and IDs.
• Existing task or status updates.
• Issues and recommendations.
• Items needing confirmation or manual Teams response.
• Safety confirmation: no Teams send/mutation actions were performed.