--- name: daily-triage description: Finds every DocJacket transaction that needs attention today and surfaces the next action per deal, as a bulleted list grouped by the urgency buckets the tool returns, most urgent first. Use when the TC asks what needs attention today, for a morning briefing, or for a triage of their active deals. license: MIT --- # Daily Triage ## When to invoke Trigger on any variation of: - "what needs my attention today?" - "give me a TC briefing" - "what's overdue?" - "what's at risk this week?" - "morning briefing" / "morning triage" - "what should I work on first?" ## Workflow **One call. The tool returns a pre-ranked, pre-bucketed feed.** 1. Call `get_next_required_actions` with `limit: 75`. (Optional: pass `transactionId` to scope to a single deal.) 2. The response is already sorted overdue-first and tagged with `urgency` per row — group by the `urgency` field as-is. That's it. **Do not fan out per transaction** — the previous v0.1 workflow called `search_transactions` then `get_open_tasks` + `get_key_dates` per active deal then sorted client-side. That's been replaced by this single call, and re-introducing the old pattern wastes 50+ tool calls per triage on a busy org. ## Output shape ``` 🔥 Overdue (3) • 1234 Main St — Inspection Deadline was due 2 days ago. • 5678 Oak Ave — Task "Order title" is 5 days overdue. • 9012 Pine Rd — Financing Deadline was due 1 day ago. ⏰ Due within 72 hours (5) • 2468 Elm — Closing Date due in 2 days. • ... 📅 Due within 2 weeks (12) • ... ``` Use the `urgency` field to decide the section header: | `urgency` | Section header | Emoji | |---|---|---| | `overdue` | Overdue | 🔥 | | `due_72h` | Due within 72 hours | ⏰ | | `due_2w` | Due within 2 weeks | 📅 | | `due_4w` | Due within 4 weeks | 📋 | Render the `rationale` string verbatim — it's already phrased for human reading. Prefix each row with the `transactionAddress` (or "—" if null) followed by an em-dash. If `summary.totalCount === 0`, render: `✅ All clear — no items need attention right now.` For each non-empty bucket, the section header should include the count from `summary` (e.g. `🔥 Overdue (3)`). The `summary` counts cover every qualifying item, not only the rows in `actions`. When `summary.truncated` is true, say the list shows the first `summary.returned` of `summary.totalCount` items (and "at least" when `summary.countsAreLowerBound` is true). ## Anti-patterns (do not do) - **Don't** call `search_transactions` + `get_open_tasks` + `get_key_dates` separately and re-sort. That was the v0.1 workflow; `get_next_required_actions` does it server-side now in one shot. - **Don't** call `get_transaction` to look up the address — `transactionAddress` is already on each row. - **Don't** call `get_next_required_actions` per transaction in a loop. One org-wide call covers everything; the `transactionId` arg is only for the rare "scope to this deal" case. - **Don't** summarize what was already completed or approved. Triage is about what still needs a decision. - **Don't** propose any writes from this skill. Triage is read-only. If the user follows up with "draft the emails" or "send these", that's a separate skill invocation (Phase 8+). - **Don't** flatten the buckets into one mixed list — the urgency grouping is the whole point. ## Example invocation trace User: _"What needs my attention today?"_ ``` [1] get_next_required_actions(limit: 75) → 26 actions returned (1 overdue, 5 due_72h, 12 due_2w, 8 due_4w) [2] Group by urgency.field; render grouped list per the Output shape table above. ``` Total: 1 tool call. Typical latency ~150ms. ## Skill version `v0.2.0` — collapsed from 4-call workflow to single `get_next_required_actions` call. The substrate now provides the ranking + rationales; the agent's job is just to render. `v0.1.0` (prior) — orchestrated `search_transactions` + `get_open_tasks` + `get_key_dates` per active transaction, then client-side bucketed + sorted. Retired 2026-05-18 when the bundled-judgment tool shipped.