Report agent
It reads the ledger, notes, pipeline, meetings, and source files, then returns what changed, what is behind, what is unsupported, and what needs a decision.
Summary
Runs
KPI memory, weekly reviews, business reviews, dashboard snapshots, pipeline reads, portfolio views, and claim checks
Reads
KPI ledger, daily notes, pipeline, meetings, decisions, client goals, reports, and source files
Writes
Ledger rows, reviews, dashboards, scorecards, report drafts, fact checks, and next-move recommendations
Guardrail
No guessed numbers. Unsupported claims are marked as stale, inferred, or unverifiable
Where it helps
The first agent should have a clear job, known inputs, and a review path the team can keep using.
What it can run
These capabilities come from the public Conversion Skills patterns. The names under each card are the documented skills the build can draw from.
The memory skill appends metric rows, records decisions, captures lessons, defines glossary terms, and retrieves the latest known value before a claim is made.
It can roll up the week, lead with what is behind, name wins and drag, read metric movement, and propose next-week priorities.
It can turn the KPI ledger into a monthly or quarterly scorecard with movement by area, an evidence-linked narrative, and three next moves.
It can render a self-contained HTML executive dashboard from the ledger, pipeline, decisions, meetings, and daily priorities. The page is a point-in-time snapshot.
It can pull pipeline health, rep performance, client risk, and owner-only portfolio views into reports with source paths and confidence notes.
It can verify factual and numeric claims against sources, assign a verdict, quote the evidence, and mark unsupported claims instead of smoothing them over.
How it works
Step 01
The agent starts from the ledger, decisions, daily notes, meetings, pipeline, client goals, and source files. If a number is missing, it says so.
Step 02
It writes the weekly review, business review, dashboard, claim check, or scorecard with source paths and confidence where the skill requires it.
Step 03
The report ends with the few things a person should decide next, instead of hiding the work behind a busy dashboard.
Human gates
The agent prepares work and keeps the trail clean. People still approve the moves that can affect buyers, customers, money, permissions, or public claims.
Outputs
The useful output is not a magic answer. It is a reviewable artifact, a clearer next move, and a source trail for anything factual.
Append-only KPI memory rows
Weekly reviews with what is behind, wins, drag, movement, and Top 3 candidates
Monthly or quarterly business reviews
Self-contained executive dashboard snapshots
Sales, portfolio, and client-health roll-ups
Fact-check tables with verdicts, evidence, and confidence
Related agents
Most useful AI systems start with one agent, then connect the next repeated job once the first one earns trust.
Public standard
Conversion Skills is the public reference for these patterns: plain files, documented skills, proof-aware outputs, and review gates. The agent pages show how that public method turns into useful business systems.