Client reports your AI drafts — with every number traced to source.
Three agent skills that turn raw KPI exports and approved client notes into evidence-linked report drafts, checked against your brand and ready for your team’s sign-off. They run in the Claude account you already use.
Organic sessions grew to 14,776R14, up 18.4%R14·R10 on August. Paid search CPA fell to $23.10R16·R12, inside the target agreed in the client’s 2 Sept notesN03.
One workflow, three skills, from raw export to signed-off report.
Start from one agreed CSV schema and the client notes you’ve approved. Each skill does one job, hands a checkable file to the next, and leaves the final call to a person on your team.
Your exports (.csv) + approved notes01 Normalize02 Draft with evidence03 Check numbers + brandYou sign off
ASK-01v1.0 · release candidate
KPI Normalizer
Turns each platform’s export into one agreed CSV schema, so every report starts from the same clean table.
In
Raw CSV exports
Out
kpi_normalized.csv + anomaly log
Maps columns to the pack schema
Flags missing periods, duplicates and unit mismatches
Curated skill catalogs already exist. We compete on documented improvement over the agent on its own and over free alternatives: fewer corrections, traceable numbers, and compatibility we keep maintaining.
How the Reporting Pack compares
For each report
Agent alone
Free skills
Reporting Pack
One agreed data schema
—
Varies
Yes
Every figure linked to source
—
Rarely
Yes
Numbers + brand QA
Manual
Manual
Built in
Readable packages + local tests
—
Varies
Every release
Maintained as models change
—
Rarely
90 days included
Commercial license for client work
n/a
Check each
Yes
Benchmark · in testing
Our target: half the preparation and review time — with no drop in quality.
We publish measured results here only after independent testing. Until then, this panel stays honest.
Corrections per report[RESULT PENDING]
Figures traceable to source[RESULT PENDING]
Time to an approved draft[RESULT PENDING]
§03 — How every release is tested
Readable. Testable. Disclosed. Reversible.
Skills run code and touch client numbers, so we treat each release like software you’d let near a client deliverable — because that’s what it is.
01
Reviewed
Every package includes its source and a test record. Independent review is pending for this release candidate.
02
Sandbox-tested
Local scripts are tested on sample inputs. End-to-end model configuration testing is pending.
03
Permissions disclosed
Each skill lists what it reads, writes and runs — on its spec page, before you pay.
04
Reversible
Every version you’ve bought stays in your library, with step-by-step rollback guidance.
Your client files never touch our store.
Skills run inside your own AI account, on your own usage. We sell the workflow; your clients’ data stays exactly where it already lives.
PERMISSIONS · ASK-02as disclosed on the spec page
Reads
kpi_normalized.csv, approved client notes you upload
Writes
A draft and an evidence index, in your session only
Runs
Bundled Python scripts, readable in the package
Network
None required
Sends
Nothing — a person on your team signs off
§04 — What you download
Real packages you can open, read and test.
No black boxes. Every skill ships as a versioned folder with its instructions, scripts, worked examples and the tests we ran.
1
Download from your libraryEvery version is kept and labelled, so you always know what your team is running.
2
Add the skills to ClaudeINSTALL.md covers setup and local verification, step by step.
3
Run the bundled exampleCompare the result with examples/expected_normalized.csv before any client data goes in.
4
Swap in this month’s exportReview the draft, work the checklist, sign it off.
Start with an AI tool that supports the Agent Skills format. Independent Claude configuration review is pending; run the included examples before client work.
Do you ever see our client data?
No. Skills run in your own AI account. The store holds your purchases and downloads — never client files.
What happens when the model changes?
We re-test and ship a new version during your 90-day update window. Older versions stay in your library if you need to roll back.
Does someone need to be technical?
If your team can export a CSV and upload a file to Claude, you can run the pack. INSTALL.md covers the rest.
Will it send reports to clients?
Never. It produces drafts and a checklist. A person on your team approves every report.
What if it isn’t right for us?
Refund policy pending merchant approval. Start with the free diagnostic to see where your time goes before you buy.
Start free · ASK-00
Find out what one client report really costs you.
Run the free diagnostic on last month’s report. You’ll get a scorecard of the hours and corrections behind it — useful whether or not you buy.