stakeholder-intelligence-brief

name	stakeholder-intelligence-brief
description	Prepare a two-minute stakeholder briefing from recent email and Teams history. Reviews the last 90 days of communications with a specific person and produces a structured snapshot of commitments made, open/unresolved items, and observable communication patterns — based only on visible signals, no inferred intent. Use when user asks: "prep me for my meeting with [person]", "brief me on [person]", "stakeholder context for [person]", "what's my history with [person]", "relationship history with [person]", "get me up to speed on [person] before our meeting", "what have I committed to [person]", "what's open with [person]", "stakeholder briefing", "two-minute brief on [person]". Do NOT use for: scheduling meetings (use schedule-meeting), preparing for a specific meeting's agenda from a transcript (use meeting-intel), or drafting messages to the person (use stakeholder-comms).
cowork	{"category":"productivity","icon":"PersonInfo"}

Stakeholder Intelligence Brief

Produce a concise, evidence-based briefing on a single stakeholder using only the user's own email and Teams history from the last 90 days.

Workflow

1. Resolve the person

• If no name was provided, ask once: "Who would you like the briefing on?"
• Resolve the name via SearchPeople / GetUserDetails.
• If multiple plausible matches exist (common name, multiple directory hits), list the top candidates with job title and email and ask the user to pick. Do not guess.

2. Gather communications (in parallel)

Run these lookups concurrently — they are independent:

• SearchM365(sources=["email"], from_user="<resolved email>", after="<today - 90d>")
• SearchM365(sources=["email"], query="to:<resolved email>", after="<today - 90d>") — for messages the user sent to them
• ListChatMessages(person="<resolved name or email>", since="<today - 90d>", top=50)
• Optionally SearchM365(sources=["teams"], from_user="<resolved email>", after="<today - 90d>") for channel mentions

Paginate with next_link if the user has dense history with this person.

3. Extract signals

From the retrieved messages, identify:

• Commitments made: things the user said they would do ("I'll send…", "I'll get back to you on…", "we'll have that by…"). Cite the message and date.
• Open asks / unresolved threads: questions or asks from the stakeholder that have no visible reply, or threads where the user said "I'll follow up" without a later resolution.
• Incomplete follow-ups: items where the user promised a deliverable whose due date has passed with no closing message.

4. Observe communication patterns

Report only what is directly visible:

• Frequency: rough cadence over the 90-day window (e.g., "weekly", "burst in March then silent")
• Response gaps: longest stretch without exchange, and any messages from the stakeholder still unanswered
• Urgency / escalation language: presence (not interpretation) of phrases like "urgent", "ASAP", "escalating", "please advise". Quote them; do not infer mood.

Do NOT speculate about intent, sentiment, or relationship health beyond the observable signals.

5. Handle thin data

If fewer than ~5 messages exist across the 90-day window, say so explicitly and summarize only what is present. Do not pad the briefing.

6. Produce the briefing

Output as plain markdown (inline, not a file) with these sections:

• Relationship Snapshot — 1-2 sentences: role, how they relate to the user, rough cadence.
• Commitments Made — bulleted list with date and message link [msg_N].
• Open or Unresolved Items — bulleted list with date and link.
• Communication Pattern Signals — frequency, gaps, escalation language (if present).

Use bracketed aliases ([msg_3], [chat_7]) so each citation is clickable.

When NOT to Use

• Drafting a message to the person → use stakeholder-comms
• Preparing a specific meeting from a transcript → use meeting-intel
• Scheduling or rescheduling time with the person → use schedule-meeting
• Performance evaluation of the person → refuse (org policy)

Guardrails

• Ground every claim in a retrieved message; never fabricate commitments, dates, or quotes.
• Quote escalation language verbatim; do not characterize tone.
• If the user is the stakeholder's manager or vice versa, still report only observable signals — no performance judgments.
• Output is inline markdown unless the user explicitly asks for a file.