<quick_start> Trigger: "business pulse", "how are we doing", "weekly numbers", "standup brief". One call gets you 90% of it:

weekly_brief({})

Then synthesize into the Output shape below (never dump raw JSON). For the new SDR cohort or deeper cuts, use the query_dataset calls in "How to run it". Reference material: reference/ in this skill

epiphan-ai-mcp-guide-skill (golden defaults, verticals pack). Owner IDs and SDR seats come from ~/.claude/skills/epiphan-house-rules-skill/reference/identifiers.md — do not restate them here. Empty-data guard: if weekly_brief errors or returns no rows, say so in the output line (NO LIVE DATA — <stage>) and never fill the gap from memory. Deliver inline; headless runs also save the brief (see "Delivery"). </quick_start>

How to run it

One-shot brief (start here):
```
weekly_brief({})
```
Returns revenue (week/MTD/QTD/YTD + prior-year), deals won/lost, new contacts by lifecycle + source, BDR activity, and AE pipeline by stage — already tuned to FY2026 (start 2025-11-01), $19.5M target, and the tool's default BDR owner set (Tim / Ron).

Track the new SDR cohort (Edgar / Vasil / Nyasha — onboarded June 2026) explicitly, passing the three SDR seat owner ids from identifiers.md ("SDR seats" table):
```
weekly_brief({ bdr_owner_ids: [<SDR seat ids>] })
query_dataset({ dataset: "rep_activity", group_by: ["owner"],
                filters: { owner_ids: [<SDR seat ids>] } })
```

Deeper cuts as needed:

query_dataset({ dataset: "pipeline_open", group_by: ["owner","stage"] })
query_dataset({ dataset: "revenue", group_by: ["period_month"], date_from: "2025-11-01", date_to: "2026-11-01" })
query_dataset({ dataset: "deals_closed", group_by: ["outcome"], date_from: <quarter start>, date_to: <quarter end> })

Synthesize into the output shape below — never just dump JSON.
Empty-data guard (before synthesis):
- weekly_brief error (auth, timeout, rate limit) → retry once. Still failing → fall back to the query_dataset calls in step 2 for each section you can; mark the run partial.
- A section with no rows (or query_dataset also failing) → print that line as NO LIVE DATA — <section> (<reason>). Never substitute a remembered, prior-run, or estimated figure.
- No section pulled live at all → output only the header + NO LIVE DATA lines, status error, and tell Tim the MCP is down. A brief with fabricated numbers is worse than no brief.

Delivery

Interactive (default): render the BUSINESS PULSE block inline in the chat.
Headless / scheduled / standup consumer: additionally (a) save the block to ~/.claude/skill-analytics/reports/business-pulse-<week-ending>.md and (b) stage a Gmail draft to tkipper@epiphan.com (from the same address) with subject BUSINESS PULSE — week ending <date> so it survives the session. Draft only — never send (Gate D). Slack DM to Tim when a Slack connector is available; otherwise the draft is the alert channel.

Output shape

BUSINESS PULSE — week ending <date>

PACE        YTD $<x> of $19.5M target  →  <n>% of FY pace   [On / Behind / Ahead]
REVENUE     Week $<x> · MTD $<x> · QTD $<x> · YoY <±%>
PIPELINE    $<x> open across <n> deals
            by stage: Discovery $<x> · … · Commit $<x>
WON / LOST  Won <n> ($<x>) · Lost <n> ($<x>) this period
BDR         Tim: <dials/connects/meetings> · Ron: <…>
TOP MOVERS  <2-3 notable deals advancing or slipping>

SO WHAT (coaching)
 1. <where the gap is, and the single highest-leverage action>
 2. <a stage with stalled volume → who multi-threads it>
 3. <a vertical/segment signal worth a play this week>

Vertical-awareness — and its honest limit

There is no native vertical/industry dimension in the datasets (group_by = period/owner/stage/outcome/country/pipeline/lifecycle/source). So any HigherEd / Community College / Live Events / Corporate / Broadcast split is derived — proxy via source/pipeline/country, or enrich the top open deals account-by-account (sales_brief). Always label a vertical breakdown as approximate; never present a fabricated clean split. (Vertical pack: epiphan-ai-mcp-guide-skill/reference/verticals.md.)

Coaching lens (tie back to the data we have)

Every 2026 loss was single-threaded below Manager level → if a big deal sits in Discovery, the play is "thread up to the economic buyer," not "follow up."
22 of 29 HigherEd deals stalled in Discovery → Discovery-stage bulk is the leading indicator to act on.

Guardrail — internal only

Output is internal. Naming partners/competitors here is fine; anything buyer-facing must stay clean (no AV-matrix mechanism, no third-party brand names — "your CMS / LMS"). The pooled figure is "a more affordable path, a starting point, not a quote." See epiphan-ai-mcp-guide-skill.

<success_criteria>

Pulled LIVE data via weekly_brief / query_dataset this run — no stale or fabricated numbers
Output follows the BUSINESS PULSE shape: pace vs $19.5M target, revenue (week/MTD/QTD/YoY), pipeline by stage, won/lost, BDR activity, top movers
Exactly three "SO WHAT" coaching takeaways, each tied to a number in the brief
Any vertical breakdown is labeled approximate (derived via source/pipeline/country proxy — no native vertical dimension)
Output stays internal-only; buyer-facing guardrails respected
Any section that could not be pulled live is printed as NO LIVE DATA — <section> (never filled from memory); run marked partial/error accordingly
Headless runs saved to ~/.claude/skill-analytics/reports/ and staged as a Gmail draft to Tim
Outcome sidecar + run-log line written </success_criteria>

<self_healing> partial = brief rendered but at least one section is NO LIVE DATA (weekly_brief degraded to query_dataset, or a dataset returned nothing) or delivery failed (file/draft not written, brief shown inline only). error = nothing pulled live this run — header + NO LIVE DATA lines only. Ladder per ~/.claude/skills/epiphan-house-rules-skill/reference/self-healing.md: retry once → degrade (query_dataset per section, flagged) → alert Tim (Gmail draft / Slack DM) → halt rather than guess. Append one line per run to ~/.claude/skill-runs/business-pulse-skill.jsonl. </self_healing>

Emit Outcome Sidecar

As the final step, write to ~/.claude/skill-analytics/last-outcome-business-pulse-skill.json:

{"ts":"[UTC ISO8601]","skill":"business-pulse-skill","version":"<frontmatter version>","variant":"default",
 "status":"[success|partial|error]","runtime_ms":[estimated ms from start],
 "metrics":{"sections_live":[n],"sections_no_data":[n],"deals_in_pipeline":[n],"coaching_takeaways":[n],"delivered_via":"inline|file|draft"},
 "error":null,"session_id":"[YYYY-MM-DD]"}

Use status "partial" if some sections were not pulled live but a brief was produced; error must name the section. Use "error" only if no live data was pulled at all.