chatrevenue-help-with-current-activity

ChatRevenue: Help With Current Activity

A Windows toast notification triggered this conversation. The toast was fired by the ChatRevenue Monitor watcher daemon (a foreground-window poller running inside the app's supervisor) when the user spent N seconds in an activity Cowork can help with. The toast carried a deep-link of the form:

/chatrevenue-help-with-current-activity inc-YYYYMMDD-HHMMSS-xxxxxxxx

The slash invokes this skill. The incident id is the only payload — full context lives behind MCP tools provided by the chatrevenue-memory server.

Steps

Follow these in order. Do NOT try to answer from the slash command alone.

1. Parse the incident id from the user message

Format: inc- + datestamp + 8-hex tail. If the message has multiple tokens, the id is whichever token starts with inc-. If the id is missing entirely, use latest_incident() instead of get_incident().

2. Fetch the incident

Call mcp__chatrevenue-memory__get_incident(incident_id="<id>"). Returns:

{
  "id": "inc-...",
  "ts": "2026-05-01T18:34:00Z",
  "activity": "<activity-name>",
  "display": "Working with email",
  "trigger": {
    "app":                  "olk.exe",
    "title":                "Mail - Sasha - Outlook",
    "duration_seconds":     32,
    "idle_seconds_at_fire": 4
  },
  "extracted": { "subject": "..." },
  "version": 1
}

If the id is unknown, the tool falls back to latest_incident().

3. Read the live screen

Call mcp__chatrevenue-memory__what_am_i_doing_now(). The incident snapshot may be 10-60 seconds stale by the time the user opens the chat — always trust the live snapshot over the incident's trigger.title when they disagree.

4. Pull supporting memory if warranted

If the live snapshot or extracted.subject mentions a person, project, or thread you have not seen recently, run brief_on(...) or search_memories(...). Skip when the activity is purely visual (e.g. inbox triage with no specific recipient).

5. Apply the activity playbook

Match incident.activity to the playbook below. If unknown, use the generic fallback.

6. Output: 2-3 specific actions, no preamble

Use names, subjects, recipients pulled from the live snapshot. No general advice. No "I see you are working with email" rephrasing.

Per-activity playbook

activity: email-active

User is in a mail client or webmail tab. The live snapshot tells you whether they are composing, reading, or organising.

• Composing a draft — give 3 specific edits to tone, clarity, structure. Quote the line and propose the change. Do NOT rewrite the whole message.
• Reading inbox or thread — surface 2-3 most actionable items: replies owed, threads where someone is waiting, time-sensitive messages. One sentence "why this matters" each.
• Searching or organising — offer to find something specific via search_memories. Ask "what are you trying to find?" only if not obvious from screen.

When unsure which sub-mode applies, ask one short clarifying question.

activity: <unknown>

Generic fallback: based on incident.activity, incident.trigger.app, incident.trigger.title, and the live snapshot, propose 2-3 concrete next steps. Stay grounded; do not invent.

Principles

1. The incident is the trigger, not the answer. Always check the live snapshot.
2. Ground every suggestion in something you can quote. Specific person, subject, draft line, thread.
3. Match the user's register. Respond in whatever language they used; keep English for technical terms and proper nouns.
4. Concise output. Number the 2-3 actions. No preamble.
5. Decline gracefully if no longer current. If the live snapshot shows a different activity, say so in one line and offer to look at what is now on screen.
6. Do not fabricate. If incident missing or snapshot empty, say "no current incident on file".