Executive reporting

A reliable Monday-morning narrative from a dozen tools without anyone manually copy-pasting numbers. The agent pulls KPIs, computes deltas, drafts a 1-page brief grounded in cited data, and emails it before the standup.

11ReportingExecutive reporting

Context Graph tribal knowledge

KPI definitions · trailing snapshots · exec tone style

Sources

GitHub

HubSpot

Grafana

Mon 06:00

Agentcy

Reporting agent

1-page brief

Output

PDF

+ email

At a glance

Inputs: GitHub (velocity, PR throughput), HubSpot (pipeline value, conversion), Stripe (revenue, churn), Grafana (uptime, latency), plus any custom KPIs via SQL.
Trigger: cron Monday 06:00 (output ready for 09:00 standup).
Output: Markdown report → PDF, email to leadership, summary in Slack #exec-brief.
Gates: read-only — every claim must cite its source row/dashboard/PR.

Stack

Layer	What we use
Triggers	Cron weekly Monday 06:00
Connectors	GitHub, HubSpot, Grafana, SQL (Stripe replica or warehouse), Google Workspace (Drive output)
Agent	Reporting agent with skills for executive-tone writing + KPI math
Output	PDF artifact written to Drive, Slack digest, optional email
Realm	`exec` — separate from engineering data so PII filters can apply

What you'll build

A weekly task that queries each KPI source.
The agent computes WoW, MoM, and trailing-4-week deltas.
A 1-page narrative is drafted: top 3 wins, top 3 risks, what changed since last week.
Every claim has a citation — link back to the dashboard, PR, deal, or query.
The PDF is rendered, dropped in Drive, and announced in Slack.

Prerequisites

Each KPI source connector enabled
A Google Drive folder for archived reports
Realm exec configured
Optional: a "tone" skill describing your CEO's preferred style (terse, data-first, no fluff)

Worked example

rego

# policies/reporting.rego
package agentcy

# Strict read-only — never write to source systems.
default allow := false

allow if {
    input.tool in {
      "github.list_pulls",
      "github.list_releases",
      "hubspot.search_deals",
      "hubspot.get_pipeline",
      "sql.run_query",
      "grafana.query",
      "grafana.render_panel",
    }
}

# Allow writing only to the designated Drive folder.
allow if {
    input.tool == "google_drive.upload"
    input.args.parent_folder == "exec-briefs"
}

Task:

yaml

name: weekly-exec-brief
schedule: "0 6 * * MON"
realm: exec
agent: reporting-agent
enabled_skills: [executive-tone, kpi-math]
prompt: |
  Draft this week's executive brief. Cover:
    1. Engineering velocity (PRs merged, releases, p95 latency from Grafana).
    2. Pipeline (HubSpot deals by stage, value, conversion vs last week).
    3. Revenue (Stripe MRR, ARR, churn, top 5 new customers).
    4. Risks (any KPI outside ±2σ of trailing 4 weeks).
  Every claim must cite a source. Keep it 1 page.
  Write the markdown to Drive folder "exec-briefs", render PDF, post a 5-line summary to #exec-brief.

What good looks like

Monday 06:00 — Exec brief — Week 17
📈 Wins
Revenue +8.2% WoW ($1.42M → $1.54M) — driven by 2 enterprise expansions ([HubSpot →])
p99 checkout latency 280ms → 190ms after the index rollout ([Grafana →])
47 PRs merged, 12 above team baseline ([GitHub →])
🔻 Risks
Pipeline coverage 2.1× → 1.7× (below 2× target) ([HubSpot →])
One enterprise deal slipping ($340k, was committed Q2) ([HubSpot deal →])
🔮 Watch
PR cycle time +18% WoW — investigate if review SLA needs attention
Full PDF →

Variations

Per-function briefs — same recipe, different KPIs and audiences (Sales brief, Engineering brief, Finance brief). Same data file in recipes.ts lists them as separate entries.
Daily sentinel — strip down to 5 KPIs, fire daily, only Slack-post on threshold breach.
Open-mic pre-meeting — 30 minutes before all-hands, post the brief into the meeting Slack channel as a kickoff document.

Troubleshooting

Numbers don't match BI tool. The brief queries the source of truth (Stripe API, HubSpot API), not your warehouse. Either reconcile, or point the recipe at the warehouse via SQL.
Citations are stale. Make sure connector caches are sync'd Sunday night before the Monday cron.
Report is too long. Add a max_words: 400 instruction to the prompt — the agent respects it.

Concept: Skills — for the tone/style skill
Connector: Grafana — embedded panel rendering
Connector: HubSpot

Executive reporting ​

At a glance ​

Stack ​

What you'll build ​

Prerequisites ​

Worked example ​

What good looks like ​

Variations ​

Troubleshooting ​

Next ​